From 5cfef813a2320e02e088ef2b3ce027953dabf7eb Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:27:39 +0000 Subject: [PATCH 01/15] release rules 15 (the version bump as the release branch's first commit, gated); release 0.3.22 section 11: the 0.3.23 cut at 6f234587 with the fold and the version miss, the 0.3.22 Windows take 1 fault and take 2, the Devnet 3 passes, the LG-4 shape on PC 2 Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 8 ++++++++ docs/plans/release-rules.md | 1 + 2 files changed, 9 insertions(+) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index 809761b38..37e612f63 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -112,3 +112,11 @@ Three commits on 96161037: bd710a36 the sub-version 3 re-pin (byte 7, igneum-pow **The steward's two items before the public repository opens, both closed:** tools/exec-attacks/lib/common.mjs MINER_KEY is the public Anvil/Hardhat default account 1 key (five harnesses use it; balance 0 and nonce 0 on both chains at 20:55 and 20:59 BST; the fix is a chain-id guard in the five harnesses and a README line, a 0.3.24 steward row); app/igneum-wallet/src/hd.rs's KEY in one history commit is the public Hardhat default account 0 key in a removed test. **Proven share, the second hour (20:59 BST):** cumulative 0.468 (1,848 of 3,952 shards since DAA 0); the hour's own segments 0.84 (1,094 of 1,304), the shortfall mostly the 34a2dbaa sweep's one-minute node restart per prover; eleven provers (five 3060s at about 85 s a segment, two 3090s, three 4090s, one A5000), a twelfth starting; the next hour's own segments should read one and the founder's 24 hours count from the first that does. **The pool hour** counts from 20:33:28 BST (dn3-pool-b's first OPEN SHARE on the 017e7037 daemons a357c581; WRONG HASH 0 since), the pool split's condition. + +## 11. The 0.3.23 cut (21:1x to 21:3x BST): the fold, the version miss, the LG-4 shape + +**release-0.3.23 = 05c2ddfe on the mirror:** 2f27ecb1 (the rights step 332b82eb with the deferred rule and the WebView2 right, the unattended driver install 50cdb8e4) + check-labels-23 e6f9ef0b + ota-token-23 757d7870 + the Windows pin to 2720d8d2 + dpi-23 ad407b13 (per-monitor DPI; BUILD-APP.bat, host.rc, host.manifest, host.cpp) + network-23 b18a5b0f (the first-run network step: the manifest's default_network, the testnet card present and refused until testnet_open, `--testnet --netsuffix=1` with the seeds as --addpeer; view.test.mjs resolved as the union of the check-labels and network tests) + install-close-23 5140e6fe (main's fold inside 15 minutes: the installer's stop step ends the window host first by path and waits up to 30 s for the unlock, CloseApplicationsFilter and SetupMutex in the .iss, install-running.flag holds every relaunch, the engine prints EXIT update and the host ends instead of restarting; tests known-failed first on tonight's PC 2 sequence) + the version bump 05c2ddfe. App gate GREEN on build-1 at every step (last 284 + 35 + 8), UI 86, pre-push 59. The version miss (rule 15): the branch carried 0.3.22 in every place until 21:26 BST, caught by the build-server lane before the host job; the exes and kit from 1c5323be are void. The Mac DMG re-cut under the lock on the 2720d8d2 Mac pair (igneumd efff01cd, igneum-miner 45c4c68f); the 0.3.23 host from the 05c2ddfe kit in the PC 1 slot after the hash lane's floor grid; the manifest carries default_network devnet-3 and keeps the floor file; the 0.3.23 Windows smoke is the first read of install-close-23 over a running app (the installer's own stop step, no pre-stop by the job) and of the OTA-return line through the app's own path. + +**The 0.3.22 Windows smoke, take 1 FAIL (20:54 BST), the job shape:** the installer 5ac3dc62 built clean on PC 2 (every exe 0.3.22, the MSVC host 4bc25b7f through the host gate), but the job's silent install over the running 0.3.21 app sent api/quit, the engine stopped, the window host (the same pid since 19:14 BST) never stopped and its restart ladder relaunched the old engine 10 s later, Inno could not replace the locked host with its message box suppressed, every file stayed 0.3.21. Take 2 in the 0.3.10 smoke shape (the job stops the app by pid first, installs, asserts every file's version and sha, starts the host, reads the app alive at 60 s on 0.3.22 with stdin open). The defect is install-close-23 in 0.3.23. Devnet 3: the 34a2dbaa pass complete on every node by 21:17 BST (thirty-two read-backs incl. hub-1's and pool-1's second nodes); the 2720d8d2 pass started 21:17:23 BST with dn3-g1, the pair whole (igneumd df6476ed, miner dfdc6883, paired: program id 571131ccbd6e0de9 equal with c29f33bb at epoch 2). + +**LG-4 (the founder: "use PC 1 or PC 2 or get another machine"; main's ruling):** on PC 2 tonight after both smokes, a fresh Windows user account created by job (no Igneum state, no card cache), the three timed steps (download and install, sync to the tip, dataset build to the first accepted share) as FIRST-SHARE lines, ten runs with the app uninstalled and the profile wiped between runs, no click anywhere, the rows on /evidence labelled "PC 2, fresh user account, not a fresh image, 7 October 2026", the 9-of-10 under-10-minutes bar read against them; macOS the same way as a fresh user account on the Mac tomorrow; a rented Windows VM only if the account shape fails the bar for a reason an image would change. PC 1 stays the founder's desk. diff --git a/docs/plans/release-rules.md b/docs/plans/release-rules.md index cf01dd793..e59e3a610 100644 --- a/docs/plans/release-rules.md +++ b/docs/plans/release-rules.md @@ -18,3 +18,4 @@ Every cut of the Igneum Miner app and its node runs under these. The dated plan 11. **Kill by pid, never by name,** on the shared Mac; a merge worktree never checks out master. 12. **The Discord card only when every platform is live.** Live manifest changes beyond the binaries (a moved consensus floor) go out only on the founder's explicit word, staged beside the release with their digest and a one-line diff. 13. **Ship on green:** no calendar waits; when the gates are green, publish and state the clock time (UK). Checkpoints are for slips, not for waiting. +15. **The version bump is the release branch's first commit (7 October 2026, after the 0.3.23 miss).** When a release-0.3.N branch opens, its first commit moves the three version strings (app/igneum-app/Cargo.toml and Cargo.lock, app/windows/version.h; packaging/mac/app/Info.plist and the installer's AppVersion follow them), never left to the cut: release-0.3.23 opened at 4cdcab31 and carried 0.3.22 in every place until 21:26 BST, so the app exes and the window host crossed from its first closed tip read 0.3.22 and were void. Gate check: on a push to release-0.3.N the pre-push gate reads Cargo.toml's version against the branch name and goes red on a mismatch (self-test on tonight's shape first). From d6703f1ba99b93f6a462f62d93c41321ece86c28 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:29:07 +0000 Subject: [PATCH 02/15] tools/ci/release-version-check.sh (rule 15): a release-0.3.N branch reads 0.3.N in Cargo.toml, Cargo.lock and version.h from its first commit; self-test on the 0.3.23 shape; wired into the pre-push gate Co-Authored-By: Claude Fable 5.1 --- tools/ci/pre-push.sh | 2 ++ tools/ci/release-version-check.sh | 40 +++++++++++++++++++++++++++++++ 2 files changed, 42 insertions(+) create mode 100755 tools/ci/release-version-check.sh diff --git a/tools/ci/pre-push.sh b/tools/ci/pre-push.sh index d401750cb..647a203b2 100755 --- a/tools/ci/pre-push.sh +++ b/tools/ci/pre-push.sh @@ -83,6 +83,8 @@ never_push_checks() { # the third never-push class (7 October 2026, 20:5x UK): a founder name on ANY branch, because every branch went to the public host's # mirror and a branch green-stamped before the check existed carried one onto master through the deferred merge run "no founder name, personal login, earlier business or personal address in any tracked text file (the pre-public scrub; self-test first, encoded list)" bash -c 'bash tools/ci/founder-strings-check.sh --self-test && bash tools/ci/founder-strings-check.sh' + # rule 15 (7 October 2026): a release branch reads its own version in Cargo.toml, Cargo.lock and version.h from its first commit + run "release-version: a release-0.3.N branch reads 0.3.N in Cargo.toml, Cargo.lock and version.h (self-test first)" bash -c 'bash tools/ci/release-version-check.sh --self-test && bash tools/ci/release-version-check.sh' } tree_checks() { diff --git a/tools/ci/release-version-check.sh b/tools/ci/release-version-check.sh new file mode 100755 index 000000000..ff7f09cf0 --- /dev/null +++ b/tools/ci/release-version-check.sh @@ -0,0 +1,40 @@ +#!/usr/bin/env bash +# Rule 15 (7 October 2026, after the 0.3.23 miss): the version bump is a release branch's first commit. release-0.3.23 opened +# at 4cdcab31 and carried 0.3.22 in Cargo.toml, Cargo.lock and app/windows/version.h until 21:26 BST, so the app exes and the +# window host crossed from its first closed tip read 0.3.22 and were void. On a push to release-0.3.N this check reads the +# three version places against the branch name and goes red on a mismatch; a branch that is not release-* passes. +# +# tools/ci/release-version-check.sh [] # the current branch when omitted; exit 1 with the mismatch named +# tools/ci/release-version-check.sh --self-test # a release-0.3.23 tree reading 0.3.22 fails on every place; a matching tree passes; a feature branch passes +set -euo pipefail +cd "$(dirname "$0")/../.." +check() { # -> prints each mismatch, exit 1 if any + local branch="$1" toml="$2" lock="$3" vh="$4" want bad=0 + case "$branch" in release-*) want="${branch#release-}" ;; *) echo "release-version: $branch is not a release branch; nothing to check"; return 0 ;; esac + case "$want" in *.*.*) ;; *) echo "release-version: $branch is not a three-part version (rule 1)"; return 1 ;; esac + local got_toml got_lock got_h got_rc + got_toml=$(awk -F'"' '/^version = "/{print $2; exit}' "$toml") + got_lock=$(awk '/^name = "igneum-app"$/{f=1;next} f&&/^version = /{gsub(/"/,"",$3); print $3; exit}' "$lock") + got_h=$(awk -F'"' '/IGNEUM_HOST_VERSION_STR/{print $2; exit}' "$vh") + got_rc=$(awk '/IGNEUM_HOST_VERSION_RC/{print $3; exit}' "$vh" | tr -d ' ') + [ "$got_toml" = "$want" ] || { echo "release-version: $toml reads $got_toml, the branch is $branch"; bad=1; } + [ "$got_lock" = "$want" ] || { echo "release-version: $lock reads $got_lock for igneum-app, the branch is $branch"; bad=1; } + [ "$got_h" = "$want" ] || { echo "release-version: $vh IGNEUM_HOST_VERSION_STR reads $got_h, the branch is $branch"; bad=1; } + [ "$got_rc" = "$(echo "$want" | tr . ,),0" ] || { echo "release-version: $vh IGNEUM_HOST_VERSION_RC reads $got_rc, want $(echo "$want" | tr . ,),0"; bad=1; } + [ $bad = 0 ] && echo "release-version: $branch reads $want in Cargo.toml, Cargo.lock and version.h" + return $bad +} +if [ "${1:-}" = "--self-test" ]; then + T=$(mktemp -d); trap 'rm -rf "$T"' EXIT + printf '[package]\nname = "igneum-app"\nversion = "0.3.22"\n' > "$T/Cargo.toml" + printf '[[package]]\nname = "igneum-app"\nversion = "0.3.22"\n' > "$T/Cargo.lock" + printf '#define IGNEUM_HOST_VERSION_STR "0.3.22"\n#define IGNEUM_HOST_VERSION_RC 0,3,22,0\n' > "$T/version.h" + out=$(check release-0.3.23 "$T/Cargo.toml" "$T/Cargo.lock" "$T/version.h" 2>&1 || true) + [ "$(printf '%s\n' "$out" | grep -cF ' reads 0.3.22')" = 3 ] && printf '%s\n' "$out" | grep -q 'VERSION_RC reads 0,3,22,0' || { echo "self-test: the 0.3.23 miss was not caught on every place"; printf '%s\n' "$out"; exit 1; } + sed -i.bak 's/0\.3\.22/0.3.23/g; s/0,3,22,0/0,3,23,0/' "$T/Cargo.toml" "$T/Cargo.lock" "$T/version.h" + check release-0.3.23 "$T/Cargo.toml" "$T/Cargo.lock" "$T/version.h" >/dev/null || { echo "self-test: a matching tree failed"; exit 1; } + check driver-check "$T/Cargo.toml" "$T/Cargo.lock" "$T/version.h" >/dev/null || { echo "self-test: a feature branch failed"; exit 1; } + echo "self-test passed: the 0.3.23 shape fails on every place, a matching release tree passes, a feature branch passes"; exit 0 +fi +BRANCH="${1:-$(git rev-parse --abbrev-ref HEAD)}" +check "$BRANCH" app/igneum-app/Cargo.toml app/igneum-app/Cargo.lock app/windows/version.h From 5fd95997142d3f9d30ef495c7bb84b9c927dbde5 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:29:26 +0000 Subject: [PATCH 03/15] Chip model 5.10: class v5 on with the shadow at zero leaves the stored-dataset chip at 5.1x (GDDR7) to 9.1x (HBM3) per joule because the node is a farm cost, not a chip cost; the shadow stays the only lever in the model (the founder's question, 7 October 2026) Co-Authored-By: Claude Fable 5.1 --- docs/analysis/chip-model-v3.md | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) diff --git a/docs/analysis/chip-model-v3.md b/docs/analysis/chip-model-v3.md index 4c32be3c2..e98ed5e42 100644 --- a/docs/analysis/chip-model-v3.md +++ b/docs/analysis/chip-model-v3.md @@ -337,6 +337,26 @@ measurements land. - The ALU budgets of the M5 Max and the 9070 XT, their power at the hash, and the verifier's cost at N = 100,000 program ops are estimates; the program-length lever is a design item with its own measurements, not a result. +### 5.10 Class v5 on, the shadow at zero: does the state-derived dataset make the shadow unnecessary? (7 October 2026, 21:3x UK, the founder's question "we need a solution, deep research, other methods") + +The question: with class v5 (the dataset built from the chain's execution state, refreshed per window) and the latency-shadow work of class v4 set to zero (class v3 energy on every GPU), what edge does the strongest chip keep over an RTX 5090 per joule? If it were at or under about 2x the shadow could come off after v5 and the class v4 premium (145 W on a 5090 at the unlocked core, 88 W at the 1,400 MHz lock, measured 7 October 2026) would vanish. + +What class v5 changes for the chip, from `docs/design/class-v5-stored-state.md` sections 2 and 2a: the recompute chip (`f = 0`, the dataset derived on the fly from the cache) and the stateless or stale chip are removed as categories, because every item takes a leaf of the state and the leaves refresh every window. What it does NOT change: the strongest chip was never one of those. It is the `f = 1` stored-dataset chip of 5.5, a GPU's memory system without the GPU, and under class v5 it needs one thing more, the window's leaves, which section 2a.2 prices honestly: one node serves a whole farm, the leaves ship at 16.5 KB/s to 10,000 members today (45 MB per member per window over a 1 Gbit/s WAN before the state is 7,500x today's), and the rebuild on the chip is the same 32 ms per window every GPU pays. The dataset is still derived from a seed and the state, so a central node compresses everything but the state's bytes, and the chip stores the result as before. + +The arithmetic, on the model's own figures (5.3: the activate-bound ceilings, 2.0 nJ per random read on GDDR7 and 1.2 nJ on HBM3, the static and controller watts; the 5090 at 136.1 MH/s on 326 W, 0.417 MH/W; 128 loads per hash, the shadow at zero so no ALU beside the memory). The node is a desktop-class CPU with an NVMe and 32 GB at about 85 W (approximate, from memory of such machines; a full node with the EVM executor at 1 block/s), shared by a farm (100 chips: 0.85 W each) or carried by every chip (the attacker's worst case, 85 W each). + +| Chip, class v5 on, shadow at zero | MH/s (model) | W with a farm-shared node (0.85 W) | MH/W | Edge over the 5090 per joule | W with a node per chip (85 W) | Edge | +|---|---|---|---|---|---|---| +| GDDR7 `f = 1`, 16 devices (the 5090's own memory without the GPU) | 166 | 78 | 2.12 | 5.1x | 163 | 2.5x | +| HBM3 `f = 1`, one stack | 84 | 28 | 3.02 | 7.2x | 112 | 1.8x | +| HBM3 `f = 1`, eight stacks (an H100-class package) | 666 | 175 | 3.80 | 9.1x | 259 | 6.2x | + +Every figure is modelled (arithmetic on cited memory figures, approximate where 5.3 marks it); none is measured; the node's watts are an approximate from memory. + +Reading: NO. Class v5 with the shadow at zero leaves the strongest chip at 5.1x (GDDR7) to 9.1x (HBM3, eight stacks) per joule, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost; only a chip forced to carry its own node falls near 2x, and only the small ones (one HBM3 stack at 1.8x, the GDDR7 board at 2.5x), while the eight-stack package stays at 6.2x even with a node per chip. So the shadow (class v4's 100,000 ops per hash in the memory wait, which brings the chip to 2.1x at k = 1 and 3.9x on the claimed X9 core) stays the only lever in this model that reaches the memory-system chip, and the class v4 premium is the price of that lever on today's GPUs. What class v5 buys is different and real: the recompute chip and the stale chip are gone as categories, every miner must hold and follow the chain, and a chip's dataset is wrong the moment its node is. The premium itself has two measured levers tonight: the core-clock lock (57 of 145 W back on the 5090 at 1,400 MHz for 1.4 percent of rate; the knee below 1,400 is the second pass's) and the per-card tune the app lands by itself; what would remove it is a shadow whose work is cheaper per op on a GPU than on a chip core (the research lane's question: a shadow shaped for the GPU's idle datapath at low clock, or a memory-side cost the chip cannot amortise), not the state-derived dataset. + +Per tier: a miner on class v4 pays the premium and gets the 2.1x to 3.9x chip ceiling in exchange; on class v5 with the shadow kept the ceiling stays and the dataset is the chain's; on class v5 with the shadow dropped the premium goes and the ceiling returns to 5x to 9x. The decision is the founder's; this section gives the number. + ## 6. The per-day derivation (item 2) 6 October 2026, Counter ASIC 3.0 item 2, worker `derive` (`docs/plans/counter-asic-3-derivation.md`; everything From 4ed80163b7f3c66547094f4d25435a7073dbbc5d Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:30:13 +0000 Subject: [PATCH 04/15] Site on master: live reads Devnet 3 (the live-dn3 edit, cherry-picked onto master so a master deploy carries it; main 7 Oct 2026 21:2x BST, every site deploy is master only from now) and every repository link flips from github.com/igneum-network to git.igneum.network/igneum-network (the public Forgejo host, by the founder word through main at 21:2x BST; GitHub is dark). Also: the runner pool classes in remote-run.sh and infra/build-server/forgejo.sh (the git host stand-up, bracket-form greps) Co-Authored-By: Claude Fable 5.1 --- infra/build-server/forgejo.sh | 126 +++++++++++++++++++++++++++++++ infra/build-server/remote-run.sh | 29 ++++--- site/404.html | 8 +- site/address.html | 8 +- site/app.html | 8 +- site/bench.html | 10 +-- site/block.html | 8 +- site/claims.html | 10 +-- site/dev-fee.html | 8 +- site/download.html | 8 +- site/evidence.html | 8 +- site/explorer.html | 8 +- site/faucet.html | 8 +- site/income.html | 8 +- site/index.html | 12 +-- site/journey.html | 8 +- site/ledger.html | 10 +-- site/litepaper.html | 12 +-- site/live.html | 8 +- site/metamask.html | 8 +- site/miner.html | 8 +- site/miners.html | 8 +- site/provenance.html | 8 +- site/randomx.html | 10 +-- site/scenes.html | 8 +- site/wallet.html | 8 +- 26 files changed, 249 insertions(+), 114 deletions(-) create mode 100644 infra/build-server/forgejo.sh diff --git a/infra/build-server/forgejo.sh b/infra/build-server/forgejo.sh new file mode 100644 index 000000000..71c60983a --- /dev/null +++ b/infra/build-server/forgejo.sh @@ -0,0 +1,126 @@ +#!/usr/bin/env bash +# The git host (main, 7 October 2026 20:29 BST; GitHub suspended): Forgejo 9 under docker on build-1, data under /srv/git (config +# app.ini written once with generated secrets, kept on the box only), ssh on 2222 public, http 3000 on loopback behind Caddy at +# git.igneum.network (ACME), deSEC A record by hand (dns.sh pattern). Then, by hand through the API as the owner user igneum-labs +# (credentials in ~/.config/igneum/forgejo-admin on the Mac): org igneum-network, private repos igneum and spec, the Mac key on the +# owner; registration off, anonymous read off, push-to-create off. Nightly dump to build-2 by /usr/local/bin/forgejo-backup.sh +# (forgejo-backup.timer 03:30 UTC). Runs as root on the box: scp it over and run it. Idempotent. +set -euo pipefail +mkdir -p /srv/git/data /srv/git/config +chown -R 1000:1000 /srv/git +if [ ! -s /srv/git/config/app.ini ]; then +SECRET=$(head -c 48 /dev/urandom | base64 | tr -d '/+=' | head -c 48) +JWT=$(head -c 48 /dev/urandom | base64 | tr -d '/+=' | head -c 48) +LFSJWT=$(head -c 48 /dev/urandom | base64 | tr -d '/+=' | head -c 48) +cat > /srv/git/config/app.ini </dev/null +fi +ufw allow 2222/tcp >/dev/null +if ! grep -q "^git.igneum.network" /etc/caddy/Caddyfile; then +cat >> /etc/caddy/Caddyfile <<'CADDY' + +# the git host (Forgejo in docker, infra/build-server/forgejo.sh): everything proxied, nothing cached +git.igneum.network { + tls { + issuer acme + } + reverse_proxy 127.0.0.1:3000 + request_body { + max_size 2GB + } +} +CADDY +caddy fmt --overwrite /etc/caddy/Caddyfile >/dev/null 2>&1 || true +systemctl reload caddy +fi +for i in $(seq 1 30); do curl -fsS -o /dev/null http://127.0.0.1:3000/ 2>/dev/null && break; sleep 2; done +echo "forgejo container: $(docker ps --format "{{.Names}} {{.Status}}" | grep "[f]orgejo")" +echo "http 3000: $(curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:3000/)" +echo "ssh 2222: $(timeout 5 bash -c 'exec 3<>/dev/tcp/127.0.0.1/2222; head -c 40 <&3' 2>/dev/null | tr -d '\r\n' | cut -c1-40)" diff --git a/infra/build-server/remote-run.sh b/infra/build-server/remote-run.sh index ce4479859..dadb4654f 100755 --- a/infra/build-server/remote-run.sh +++ b/infra/build-server/remote-run.sh @@ -23,7 +23,7 @@ # 3. Appends one JSON line to /srv/builds/_log/builds.jsonl (the worker dashboard reads it; asked for by main on 6 October # 2026): on success, on failure and on the slot give-up. UTC ISO 8601 Z times, numbers unquoted, unknown fields omitted, # artefacts with bytes and sha256 only when the run succeeded (a failed build would list the previous build's files), -# the line kept under 4 KB. Since the evening of 6 October 2026 (the founder: "make sure we are fixing and learning from all the +# the line kept under 4 KB. Since the evening of 6 October 2026 (the project lead: "make sure we are fixing and learning from all the # errors here") the line also carries "class" and "run_log": # - PRE-FLIGHT before cargo runs: the manifest must parse (cargo metadata --no-deps) and every `-p` package must exist # (a workspace member, else cargo pkgid --offline); a refusal is exit 3, class preflight-manifest or preflight-package, @@ -320,7 +320,7 @@ PY SLOTS_DIR="$IGNEUM_BUILD_SLOTS_DIR" slots=$(cat "$SLOTS_DIR/slots" 2>/dev/null || echo 1); [ "$slots" -ge 1 ] 2>/dev/null || slots=1 -JOBS_ALONE="${JOBS_ALONE:-88}"; JOBS_SHARED="${JOBS_SHARED:-44}" # the founder, 7 Oct 2026: 88 of 96 cores, 8 reserved for the release builds, the seed and the observers +JOBS_ALONE="${JOBS_ALONE:-88}"; JOBS_SHARED="${JOBS_SHARED:-44}" # the project lead, 7 Oct 2026: 88 of 96 cores, 8 reserved for the release builds, the seed and the observers holder_line() { printf 'pid %s since %sZ waited %s s: %s\n' "$BR_PID" "$(date -u +%H:%M:%S)" "$1" "$BR_LABEL"; } give_up() { # echo "build-remote: gave up waiting for $1 after 2 h" >&2 @@ -480,18 +480,27 @@ BR_RUN_LOG="$RUN_LOG_DIR/$BR_HOST-$BR_T0-$BR_PID.log"; export BR_RUN_LOG # (nice 0, a gate or a release build) is outside the pool and keeps the full set. ncpu=$(nproc); cores_str="0-$((ncpu - 1))" pool_fds=(); pool_cores=() +# the pool's priority classes (lease.sh, main 7 Oct 2026 20:37 BST: release > v5 > measure > adv): a bounded build or suite is class +# release (rank 0) unless its label names a "v5 gate" or "v5 kit" (class v5, rank 1); the wait file carries "class /:" so +# `lease pool` sweeps (class adv) yield to it, and it yields while a higher class waits (nothing outranks release) +pool_class() { case "$BR_LABEL" in *"v5 gate"*|*"v5 kit"*) echo "v5/1" ;; *) echo "release/0" ;; esac; } pool_take() { # : lease up to free pool cores, at least , waiting up to 2 h; sets pool_cores/pool_fds - local want="$1" min="$2" t0 c fd taken + local want="$1" min="$2" t0 c fd taken cls rank higher + cls=$(pool_class); rank=${cls#*/} t0=$(date +%s) while :; do taken=0; pool_fds=(); pool_cores=() - for ((c = ${BR_POOL_RESERVE:-8}; c < ncpu && taken < want; c++)); do - exec {fd}>>"$SLOTS_DIR/core-$c" - if flock -n "$fd"; then pool_fds+=("$fd"); pool_cores+=("$c"); taken=$((taken + 1)); else exec {fd}>&-; fi - done - if [ "$taken" -ge "$min" ]; then return 0; fi - for fd in "${pool_fds[@]}"; do exec {fd}>&-; done; pool_fds=(); pool_cores=() - [ -f "$waitfile" ] || { echo "build-remote: the bounded pool has $taken free core(s) (want $want, at least $min): waiting for the pool" >&2; holder_line "$(( $(date +%s) - BR_T0 ))" > "$waitfile"; } + higher=$(cat "$SLOTS_DIR"/wait-* 2>/dev/null | grep -v "^pid $BR_PID " | sed -n 's/.* class [a-z0-9]*\/\([0-9]\):.*/\1/p' | awk -v r="$rank" '$1 < r' | head -1) + if [ -z "$higher" ]; then + for ((c = ${BR_POOL_RESERVE:-8}; c < ncpu && taken < want; c++)); do + exec {fd}>>"$SLOTS_DIR/core-$c" + if flock -n "$fd"; then pool_fds+=("$fd"); pool_cores+=("$c"); taken=$((taken + 1)); else exec {fd}>&-; fi + done + if [ "$taken" -ge "$min" ]; then return 0; fi + for fd in "${pool_fds[@]}"; do exec {fd}>&-; done; pool_fds=(); pool_cores=() + fi + [ -f "$waitfile" ] || { echo "build-remote: the bounded pool has $taken free core(s) (want $want, at least $min, class ${cls%/*}): waiting for the pool" >&2; } + printf '%s class %s: pool\n' "$(holder_line "$(( $(date +%s) - BR_T0 ))")" "$cls" > "$waitfile" [ $(( $(date +%s) - t0 )) -lt 7200 ] || give_up "the bounded pool" sleep 10 done diff --git a/site/404.html b/site/404.html index a3edb613c..41a840e3d 100644 --- a/site/404.html +++ b/site/404.html @@ -131,7 +131,7 @@ main{flex:1}
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -204,12 +204,12 @@ main{flex:1}

Mined by GPUs. Proven by fire.

WindowsmacOSLinuxHiveOS
- Specification and test vectors + Specification and test vectors

Run

diff --git a/site/address.html b/site/address.html index 81ce952d5..ab1160759 100644 --- a/site/address.html +++ b/site/address.html @@ -179,7 +179,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -259,12 +259,12 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin

Mined by GPUs. Proven by fire.

WindowsmacOSLinuxHiveOS
- Specification and test vectors + Specification and test vectors

Run

Download Ember @@ -292,7 +292,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
diff --git a/site/app.html b/site/app.html index 14e818c54..6f7d58a7d 100644 --- a/site/app.html +++ b/site/app.html @@ -153,7 +153,7 @@
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -456,12 +456,12 @@

Mined by GPUs. Proven by fire.

WindowsmacOSLinuxHiveOS
- Specification and test vectors + Specification and test vectors

Run

diff --git a/site/bench.html b/site/bench.html index 0a136a9e5..75fce501d 100644 --- a/site/bench.html +++ b/site/bench.html @@ -158,7 +158,7 @@ table{min-width:560px}
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -428,7 +428,7 @@ table{min-width:560px}
Harness run, same binariesDevnet profile60x profile (--fast-time)
tools/finality-attacks s3, dishonest aggregatorscatalogue 32 min at SCALE 0.6 on the 3 Oct node (above); on today's rule the window fills at DAA 7,200 = 20 min at 6 blocks/s before any lock113 s wall: 16 locks per node, 0 conflicting certificates, lock hashes agree, median lock latency 1,018 ms, PASS
tools/harness s3, partition and heal (in-process simulator of the devnet-v4 line)983 s wall, four cuts of 120 / 600 / 1,800 / 3,700 virtual s (46 / 151 / 400 / 831 s wall), all PASS43 s wall, three cuts of 10 / 30 / 62 virtual s (18 / 20 / 26 s wall), all PASS; the 62-s cut beyond the 60-s merge depth converged with a 34-block reorg

Both harnesses take --fast-time (tools/harness/lib/net.mjs, tools/finality-attacks/lib/net.mjs): the node and the simulator then come from vendor/igneum-node/target-integration (the file carries fields only the devnet-v4 line knows), the merge-depth scenarios scale their cuts with the profile, and the finality runs default to the --quick scale.

Linux cross-compile. infra/cross/build-linux.sh: cargo-zigbuild 0.23.4 with zig 0.17.0 (brew install zig, cargo install cargo-zigbuild, rustup target add x86_64-unknown-linux-gnu), target x86_64-unknown-linux-gnu.2.36 (Debian 12 on the servers), -p kaspad -p igneum-miner --features kaspad/igneum-pow, target dir vendor/igneum-node/target-linux. Cold build: 1,856 s (30 min 56 s) at 4 jobs, nice 19, on this loaded machine; rocksdb (librocksdb-sys C++), lz4, blst, secp256k1 and the execution layer's crates all linked through zig; no crate failed, so cross was not needed (Docker is not installed here anyway). Output: ELF x86-64 PIE, igneumd 46,883,304 B and igneum-miner 8,978,424 B, dynamically linked against libc and libm only. Verified on the seed node (Debian 12, glibc 2.36, 2 vCPU) in <server path>: igneumd --version = igneumd 2.1.0, igneum-miner --help prints its usage, and a 60-s run of the cross-compiled node on igneum-devnet-951 (the 60x profile at genesis bits 2^16, real proof of work) with the cross-compiled miner on one CPU thread: the miner adopted the node's 60-block epoch from the template, built its 256 MiB cache in 395 ms, found 7 blocks at 0.011 MH/s, all 7 accepted by the node's own PoW check, 0 rejected (the x86 build of the lottery hash agrees between miner and node; neither binary exposes a standalone vector check). Against the alternatives: the seed's own build took 55 min on its 2 vCPU (4 Oct 2026, above), the builder VM 10 to 25 min on 8 vCPU plus its creation and deletion. BIN_SOURCE=mac is wired into infra/cloud-devnet/provision.sh (no builder VM, build/bin from the cross-compile) and infra/seed-nodes/stage-v4.sh (upload to <server path>, install as before).

-

CI. .github/workflows/ci.yml runs on every push and pull request of the private repository: igneum-pow cargo test --release and the igneum-census build (49 s), the two simulators' --quick modes under a 120-s timeout (49 s for the job; sim/finality_v2.py --quick is now a true smoke run, 149 s at nice 19 on this loaded Mac and under 40 s on the runner, was 745 s; sim/difficulty/sim.py --quick is new, 36 s here), the site build, an internal link check of site/*.html (151 links, 0 broken) and a gh-free identity grep of the public export list after the generic scrub (tools/ci/identity-check.sh, tools/ci/forbidden-strings.txt: 157 files, 0 hits). First run green: https://github.com/igneum-network/igneum/actions/runs/37193811336, 54 s from trigger to completion. The node fork is gitignored and too big for the free runners today; the workflow says so.

+

CI. .github/workflows/ci.yml runs on every push and pull request of the private repository: igneum-pow cargo test --release and the igneum-census build (49 s), the two simulators' --quick modes under a 120-s timeout (49 s for the job; sim/finality_v2.py --quick is now a true smoke run, 149 s at nice 19 on this loaded Mac and under 40 s on the runner, was 745 s; sim/difficulty/sim.py --quick is new, 36 s here), the site build, an internal link check of site/*.html (151 links, 0 broken) and a gh-free identity grep of the public export list after the generic scrub (tools/ci/identity-check.sh, tools/ci/forbidden-strings.txt: 157 files, 0 hits). First run green: https://git.igneum.network/igneum-network/igneum/actions/runs/37193811336, 54 s from trigger to completion. The node fork is gitignored and too big for the free runners today; the workflow says so.

Not done: igneum-harness-sim's per-block cost on the devnet-v4 line (about 70 ms here against 3 ms on the ordering-layer branch, the execution layer's follower) is what still bounds the harness, not the clocks; the fast-time presence window floors at one checkpoint; the GPU workers were not run on fast time (the CPU miner proved the swap).

4 October 2026, first finality lock on the live devnet: checkpoint 242 at 77.4% of all weight, two hours after genesis

@@ -975,12 +975,12 @@ table{min-width:560px}

Mined by GPUs. Proven by fire.

WindowsmacOSLinuxHiveOS
- Specification and test vectors + Specification and test vectors

Run

diff --git a/site/block.html b/site/block.html index 2868295a8..036aa2e9d 100644 --- a/site/block.html +++ b/site/block.html @@ -179,7 +179,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -260,12 +260,12 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin

Mined by GPUs. Proven by fire.

WindowsmacOSLinuxHiveOS
- Specification and test vectors + Specification and test vectors

Run

Download Ember @@ -293,7 +293,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
diff --git a/site/claims.html b/site/claims.html index af4758d91..7509ce490 100644 --- a/site/claims.html +++ b/site/claims.html @@ -157,7 +157,7 @@
Download -
DiscordGitHubReddit
+
DiscordGitHubReddit
@@ -231,7 +231,7 @@
  • Finality that never pauses. No. A lock needs two thirds of all 30-day mining weight. Whenever less than two thirds of that weight is connected and signing, finality pauses until it returns or ages out of the window, up to 30 days. The chain keeps running on proof of work and the node reports the pause.
  • A finished protocol. The sustained-mining finality rule is the newest piece and the one that external review will try hardest to break. The specification, the review and the benchmarks are published as they happen.
    • -

    Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: github.com/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

    +

    Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: git.igneum.network/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

    The ledger: every criticismThe evidence pageRead it in the litepaper
    @@ -244,12 +244,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/dev-fee.html b/site/dev-fee.html index 83a8afd22..594db38e3 100644 --- a/site/dev-fee.html +++ b/site/dev-fee.html @@ -157,7 +157,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -255,12 +255,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/download.html b/site/download.html index b29b7fecb..959aa08ee 100644 --- a/site/download.html +++ b/site/download.html @@ -147,7 +147,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -294,12 +294,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/evidence.html b/site/evidence.html index c92824678..e9e4d6eb4 100644 --- a/site/evidence.html +++ b/site/evidence.html @@ -168,7 +168,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -289,12 +289,12 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -322,7 +322,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/explorer.html b/site/explorer.html index 564ed5351..c89f8f766 100644 --- a/site/explorer.html +++ b/site/explorer.html @@ -178,7 +178,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -281,12 +281,12 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -314,7 +314,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/faucet.html b/site/faucet.html index eb13569be..11d809d95 100644 --- a/site/faucet.html +++ b/site/faucet.html @@ -160,7 +160,7 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:14px;overflo
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -254,12 +254,12 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:14px;overflo

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -287,7 +287,7 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:14px;overflo Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/income.html b/site/income.html index 2f740ff16..b51e91b86 100644 --- a/site/income.html +++ b/site/income.html @@ -117,7 +117,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -222,12 +222,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/index.html b/site/index.html index f645bcf18..101adf755 100644 --- a/site/index.html +++ b/site/index.html @@ -28,7 +28,7 @@ @@ -149,7 +149,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -247,7 +247,7 @@

    The people running cards like yours.

    DiscordSay what breaks. Ask anything. - GitHubThe spec, the vectors, the issues. + GitHubThe spec, the vectors, the issues. RedditThreads from miners, answered.

    Every criticism, answered or conceded, is on the ledger, with what Igneum does not claim.

    @@ -263,12 +263,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/journey.html b/site/journey.html index 7e16e31b9..19ff4c88b 100644 --- a/site/journey.html +++ b/site/journey.html @@ -157,7 +157,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -280,12 +280,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/ledger.html b/site/ledger.html index 532e7aaec..d84f7b956 100644 --- a/site/ledger.html +++ b/site/ledger.html @@ -161,7 +161,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -1374,7 +1374,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c
    The answer as first written

    Correct. a repository file has on_chain_block and no reorg hook, so a transaction unwound by a selected-chain reorg is neither re-queued nor re-executed; the new state word reports reorged out with reorgedFrom, which tells a wallet to resend and tells nobody else. Fix: on every virtualChainChanged removal the executor hands the unwound transactions back to the pool as pending (nonce order kept, the same validity checks as a fresh submission, the 10,000-hash memory of P17 cleared on re-inclusion); a unit test that unwinds a block and sees its transactions re-queued; the conformance driver's reorged out case then ends in executed again without a resend.

    -

    Source: the project's criticism ledger, a file in the repository, rendered to this page at build time; the repository is published at the public testnet. A criticism that is not here, or that shows an entry is wrong, is added with credit if wanted: hello@igneum.network or an issue on the specification repository.

    +

    Source: the project's criticism ledger, a file in the repository, rendered to this page at build time; the repository is published at the public testnet. A criticism that is not here, or that shows an entry is wrong, is added with credit if wanted: hello@igneum.network or an issue on the specification repository.

    @@ -1385,12 +1385,12 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -1418,7 +1418,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/litepaper.html b/site/litepaper.html index b2d0903a9..813e37bd9 100644 --- a/site/litepaper.html +++ b/site/litepaper.html @@ -214,7 +214,7 @@ body.all .pager{display:none}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -467,7 +467,7 @@ body.all .pager{display:none} Changes over timeNone. A fixed design, unchanged for seven yearsA new program every hour, its memory pattern with it; era draws and reserved families on a schedule fixed at genesis. Nobody touches it Seed grindingNot applicable, the program comes from the hash inputClosed by a verifiable delay between seed and program Useful workNone. Hashing onlyEvery NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server (eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026); they sell proofs to other chains. AMD and Apple cards mine, and a prover for them lands when a zkVM ships one - Track recordAbout seven years without a shipped chip; the one announced, Bitmain’s Antminer X9, was withdrawn in May 2026 before launchZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (github.com/igneum-network/spec). The node, the miner and the wallet are in a private repository until the public testnet + Track recordAbout seven years without a shipped chip; the one announced, Bitmain’s Antminer X9, was withdrawn in May 2026 before launchZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (git.igneum.network/igneum-network/spec). The node, the miner and the wallet are in a private repository until the public testnet

    Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 the live devnet crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

    @@ -815,7 +815,7 @@ body.all .pager{display:none}
  • Finality that never pauses. No. A lock needs two thirds of all 30-day mining weight. Whenever less than two thirds of that weight is connected and signing, finality pauses until it returns or ages out of the window, up to 30 days. The chain keeps running on proof of work and the node reports the pause.
  • A finished protocol. The sustained-mining finality rule is the newest piece and the one that external review will try hardest to break. The specification, the review and the benchmarks are published as they happen.
    • -

    Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: github.com/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

    +

    Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: git.igneum.network/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

    @@ -829,12 +829,12 @@ body.all .pager{display:none}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/live.html b/site/live.html index 508385351..8d8bbc379 100644 --- a/site/live.html +++ b/site/live.html @@ -298,7 +298,7 @@ details.tablebar summary{display:flex;align-items:center}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -496,12 +496,12 @@ details.tablebar summary{display:flex;align-items:center}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -529,7 +529,7 @@ details.tablebar summary{display:flex;align-items:center} Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/metamask.html b/site/metamask.html index 1fb3be95e..5ec26a177 100644 --- a/site/metamask.html +++ b/site/metamask.html @@ -159,7 +159,7 @@ ol{margin:0 0 14px;padding-left:22px}li{margin-bottom:6px}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -287,12 +287,12 @@ ol{margin:0 0 14px;padding-left:22px}li{margin-bottom:6px}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -320,7 +320,7 @@ ol{margin:0 0 14px;padding-left:22px}li{margin-bottom:6px} Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/miner.html b/site/miner.html index 82423d770..38ff0b2e1 100644 --- a/site/miner.html +++ b/site/miner.html @@ -154,7 +154,7 @@ pre b{color:var(--molten-text);font-weight:500}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -438,12 +438,12 @@ pre b{color:var(--molten-text);font-weight:500}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    Download Ember @@ -471,7 +471,7 @@ pre b{color:var(--molten-text);font-weight:500} Downloads: devnet build Community Discord hello@igneum.network - An issue on the spec + An issue on the spec
    diff --git a/site/miners.html b/site/miners.html index de39cb44e..24f1d41df 100644 --- a/site/miners.html +++ b/site/miners.html @@ -158,7 +158,7 @@ table{min-width:560px}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -275,12 +275,12 @@ table{min-width:560px}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/provenance.html b/site/provenance.html index bf7c24b53..4c6a99d16 100644 --- a/site/provenance.html +++ b/site/provenance.html @@ -158,7 +158,7 @@ table{min-width:560px}
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -251,12 +251,12 @@ table{min-width:560px}

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/randomx.html b/site/randomx.html index f408d7396..ac82ae044 100644 --- a/site/randomx.html +++ b/site/randomx.html @@ -157,7 +157,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -230,7 +230,7 @@ Changes over timeNone. A fixed design, unchanged for seven yearsA new program every hour, its memory pattern with it; era draws and reserved families on a schedule fixed at genesis. Nobody touches it Seed grindingNot applicable, the program comes from the hash inputClosed by a verifiable delay between seed and program Useful workNone. Hashing onlyEvery NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server (eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026); they sell proofs to other chains. AMD and Apple cards mine, and a prover for them lands when a zkVM ships one - Track recordAbout seven years without a shipped chip; the one announced, Bitmain’s Antminer X9, was withdrawn in May 2026 before launchZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (github.com/igneum-network/spec). The node, the miner and the wallet are in a private repository until the public testnet + Track recordAbout seven years without a shipped chip; the one announced, Bitmain’s Antminer X9, was withdrawn in May 2026 before launchZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (git.igneum.network/igneum-network/spec). The node, the miner and the wallet are in a private repository until the public testnet

    Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 the live devnet crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

    @@ -246,12 +246,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/scenes.html b/site/scenes.html index 2a2310bdf..ffa13a5fa 100644 --- a/site/scenes.html +++ b/site/scenes.html @@ -135,7 +135,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -233,12 +233,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    diff --git a/site/wallet.html b/site/wallet.html index e4b035fc8..ff8f8be21 100644 --- a/site/wallet.html +++ b/site/wallet.html @@ -167,7 +167,7 @@
    Download -
    DiscordGitHubReddit
    +
    DiscordGitHubReddit
    @@ -381,12 +381,12 @@

    Mined by GPUs. Proven by fire.

    WindowsmacOSLinuxHiveOS
    - Specification and test vectors + Specification and test vectors

    Run

    From 83a947d274d3cbf0779c84293dad3f668c88de64 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:35:17 +0000 Subject: [PATCH 05/15] rule 15 at six places; release 0.3.22: the 0.3.22 Windows take 2 fault and the skip, the version miss's second layer, the 0.3.23 pairs Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 2 ++ docs/plans/release-rules.md | 2 +- tools/ci/release-version-check.sh | 16 +++++++++++----- 3 files changed, 14 insertions(+), 6 deletions(-) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index 37e612f63..03ef2b509 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -120,3 +120,5 @@ Three commits on 96161037: bd710a36 the sub-version 3 re-pin (byte 7, igneum-pow **The 0.3.22 Windows smoke, take 1 FAIL (20:54 BST), the job shape:** the installer 5ac3dc62 built clean on PC 2 (every exe 0.3.22, the MSVC host 4bc25b7f through the host gate), but the job's silent install over the running 0.3.21 app sent api/quit, the engine stopped, the window host (the same pid since 19:14 BST) never stopped and its restart ladder relaunched the old engine 10 s later, Inno could not replace the locked host with its message box suppressed, every file stayed 0.3.21. Take 2 in the 0.3.10 smoke shape (the job stops the app by pid first, installs, asserts every file's version and sha, starts the host, reads the app alive at 60 s on 0.3.22 with stdin open). The defect is install-close-23 in 0.3.23. Devnet 3: the 34a2dbaa pass complete on every node by 21:17 BST (thirty-two read-backs incl. hub-1's and pool-1's second nodes); the 2720d8d2 pass started 21:17:23 BST with dn3-g1, the pair whole (igneumd df6476ed, miner dfdc6883, paired: program id 571131ccbd6e0de9 equal with c29f33bb at epoch 2). **LG-4 (the founder: "use PC 1 or PC 2 or get another machine"; main's ruling):** on PC 2 tonight after both smokes, a fresh Windows user account created by job (no Igneum state, no card cache), the three timed steps (download and install, sync to the tip, dataset build to the first accepted share) as FIRST-SHARE lines, ten runs with the app uninstalled and the profile wiped between runs, no click anywhere, the rows on /evidence labelled "PC 2, fresh user account, not a fresh image, 7 October 2026", the 9-of-10 under-10-minutes bar read against them; macOS the same way as a fresh user account on the Mac tomorrow; a rented Windows VM only if the account shape fails the bar for a reason an image would change. PC 1 stays the founder's desk. + +**0.3.22 Windows take 2 FAIL (21:13 BST) and the skip:** the job's detached helper was ended with the job's process tree before its first sleep ran out (Start-Process stays in the runner's tree; a survivor needs Win32_Process.Create or a scheduled task); nothing was installed, every file still 0.3.21, the payload untouched. Ruling: no take 3; the 0.3.22 Windows entry is skipped (the Mac and HiveOS entries stand); the first install over a running app is 0.3.23's installer with install-close-23 in the simplest job shape (Start-Process -Wait; the installer's own stop step and the install-running flag do the work), its smoke read the gate line for both; the Discord card publishes on the 0.3.23 Windows entry. **The version miss, second layer (21:30 BST):** the box cross of 0.3.23's exes failed in build.rs because igneum-app.rc still read 0.3.22 (Info.plist and the installer's AppVersion too); fixed at release-0.3.23 = 7c7489ac, the rule 15 check extended to six places and green on the tree; the 0.3.23 node pairs under /srv/artefacts/0323-2720d8d2/ (hands f2cf6a87/fb147dd1, seed 3edaf83d/be5ca735, win 8326d78a/38c74545) with the engine string. diff --git a/docs/plans/release-rules.md b/docs/plans/release-rules.md index e59e3a610..ecddb7a3a 100644 --- a/docs/plans/release-rules.md +++ b/docs/plans/release-rules.md @@ -18,4 +18,4 @@ Every cut of the Igneum Miner app and its node runs under these. The dated plan 11. **Kill by pid, never by name,** on the shared Mac; a merge worktree never checks out master. 12. **The Discord card only when every platform is live.** Live manifest changes beyond the binaries (a moved consensus floor) go out only on the founder's explicit word, staged beside the release with their digest and a one-line diff. 13. **Ship on green:** no calendar waits; when the gates are green, publish and state the clock time (UK). Checkpoints are for slips, not for waiting. -15. **The version bump is the release branch's first commit (7 October 2026, after the 0.3.23 miss).** When a release-0.3.N branch opens, its first commit moves the three version strings (app/igneum-app/Cargo.toml and Cargo.lock, app/windows/version.h; packaging/mac/app/Info.plist and the installer's AppVersion follow them), never left to the cut: release-0.3.23 opened at 4cdcab31 and carried 0.3.22 in every place until 21:26 BST, so the app exes and the window host crossed from its first closed tip read 0.3.22 and were void. Gate check: on a push to release-0.3.N the pre-push gate reads Cargo.toml's version against the branch name and goes red on a mismatch (self-test on tonight's shape first). +15. **The version bump is the release branch's first commit (7 October 2026, after the 0.3.23 miss).** When a release-0.3.N branch opens, its first commit moves the six version places (app/igneum-app/Cargo.toml and Cargo.lock, app/windows/version.h, app/igneum-app/resources/igneum-app.rc's four fields, packaging/mac/app/Info.plist, the installer's AppVersion), never left to the cut: release-0.3.23 opened at 4cdcab31 and carried 0.3.22 in every place until 21:26 BST, so the app exes and the window host crossed from its first closed tip read 0.3.22 and were void. Gate check: on a push to release-0.3.N the pre-push gate (tools/ci/release-version-check.sh, self-test on tonight's shape first) reads all six places against the branch name and goes red on any mismatch; the .rc was the second layer of the same miss (the box cross failed in build.rs at 21:30 BST). diff --git a/tools/ci/release-version-check.sh b/tools/ci/release-version-check.sh index ff7f09cf0..1cb8404a9 100755 --- a/tools/ci/release-version-check.sh +++ b/tools/ci/release-version-check.sh @@ -2,14 +2,14 @@ # Rule 15 (7 October 2026, after the 0.3.23 miss): the version bump is a release branch's first commit. release-0.3.23 opened # at 4cdcab31 and carried 0.3.22 in Cargo.toml, Cargo.lock and app/windows/version.h until 21:26 BST, so the app exes and the # window host crossed from its first closed tip read 0.3.22 and were void. On a push to release-0.3.N this check reads the -# three version places against the branch name and goes red on a mismatch; a branch that is not release-* passes. +# six version places (Cargo.toml, Cargo.lock, version.h, igneum-app.rc's four fields, Info.plist, the installer's AppVersion) against the branch name and goes red on a mismatch; a branch that is not release-* passes. # # tools/ci/release-version-check.sh [] # the current branch when omitted; exit 1 with the mismatch named # tools/ci/release-version-check.sh --self-test # a release-0.3.23 tree reading 0.3.22 fails on every place; a matching tree passes; a feature branch passes set -euo pipefail cd "$(dirname "$0")/../.." -check() { # -> prints each mismatch, exit 1 if any - local branch="$1" toml="$2" lock="$3" vh="$4" want bad=0 +check() { # [ ] -> prints each mismatch, exit 1 if any + local branch="$1" toml="$2" lock="$3" vh="$4" rc="${5:-}" plist="${6:-}" iss="${7:-}" want bad=0 case "$branch" in release-*) want="${branch#release-}" ;; *) echo "release-version: $branch is not a release branch; nothing to check"; return 0 ;; esac case "$want" in *.*.*) ;; *) echo "release-version: $branch is not a three-part version (rule 1)"; return 1 ;; esac local got_toml got_lock got_h got_rc @@ -21,7 +21,13 @@ check() { # -> prints each mis [ "$got_lock" = "$want" ] || { echo "release-version: $lock reads $got_lock for igneum-app, the branch is $branch"; bad=1; } [ "$got_h" = "$want" ] || { echo "release-version: $vh IGNEUM_HOST_VERSION_STR reads $got_h, the branch is $branch"; bad=1; } [ "$got_rc" = "$(echo "$want" | tr . ,),0" ] || { echo "release-version: $vh IGNEUM_HOST_VERSION_RC reads $got_rc, want $(echo "$want" | tr . ,),0"; bad=1; } - [ $bad = 0 ] && echo "release-version: $branch reads $want in Cargo.toml, Cargo.lock and version.h" + if [ -n "$rc" ]; then + local n; n=$(grep -cE "^(FILEVERSION|PRODUCTVERSION) +$(echo "$want" | tr . ,),0\$|VALUE \"(FileVersion|ProductVersion)\", +\"$want\"" "$rc" || true) + [ "$n" = 4 ] || { echo "release-version: $rc carries $n of 4 version fields at $want (build.rs refuses the cross otherwise)"; bad=1; } + grep -qF "$want" "$plist" || { echo "release-version: $plist does not read $want"; bad=1; } + grep -qF "#define AppVersion \"$want\"" "$iss" || { echo "release-version: $iss AppVersion does not read $want"; bad=1; } + fi + [ $bad = 0 ] && echo "release-version: $branch reads $want in Cargo.toml, Cargo.lock, version.h${rc:+, igneum-app.rc, Info.plist and the installer}" return $bad } if [ "${1:-}" = "--self-test" ]; then @@ -37,4 +43,4 @@ if [ "${1:-}" = "--self-test" ]; then echo "self-test passed: the 0.3.23 shape fails on every place, a matching release tree passes, a feature branch passes"; exit 0 fi BRANCH="${1:-$(git rev-parse --abbrev-ref HEAD)}" -check "$BRANCH" app/igneum-app/Cargo.toml app/igneum-app/Cargo.lock app/windows/version.h +check "$BRANCH" app/igneum-app/Cargo.toml app/igneum-app/Cargo.lock app/windows/version.h app/igneum-app/resources/igneum-app.rc packaging/mac/app/Info.plist packaging/windows/Igneum-Miner.iss From e7e7b8b62f87177cc838afb9228f97720ee9f0cd Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:38:58 +0000 Subject: [PATCH 06/15] Counter ASIC 3.0 status: the founder's premium question and section 5.10's answer, the research lane's reading of the three shapes, the five ratios and the defender's ruling (the floor closes the hot-set half, not the class), adv-mixer-2's confirmation of AP-F4-1 Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index 7c18c5f24..344e5ca6b 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -455,7 +455,7 @@ Reading so far: the class v4 premium at the unlocked clock is +145 W on this car | 1,400 | 134.98 | 316.3 | 0.427 | 134.68 | 228.0 | 0.591 | 1,387 | | unlocked, end | 136.75 | 473.0 | | 136.5x | 329.7 | | 2,843 / 2,850 | -Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). THE FOUNDER'S QUESTION ON THE CLASS V4 PREMIUM ("we need a solution, deep research, other methods, something must be doable even if it is revolutionary"; a research lane, counter-asic-4, on the literature and the alternatives by 23:30Z). THE ONE COMPUTATION FROM THIS LANE (chip-model-v3 section 5.10, on the mirror's master at aa829826): class v5 ON and the shadow at ZERO leaves the strongest chip, the f = 1 stored-dataset chip, at 5.1x (GDDR7, the 5090's own 16 devices without the GPU: 166 MH/s at 78 W with a farm-shared node) to 9.1x (HBM3 eight stacks: 666 MH/s at 175 W) per joule over the 5090's 0.417 MH/W, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost (class-v5 2a.2: one node serves a farm, the leaves ship at 16.5 KB/s to 10,000 members, the rebuild is the same 32 ms per window every GPU pays); only a chip forced to carry its own node (85 W, approximate) falls near 2x, and only the small ones (one HBM3 stack 1.8x, the GDDR7 board 2.5x; the eight-stack package 6.2x). So the shadow stays the only lever in the model that reaches the memory-system chip, and the premium is its price; what class v5 buys is the recompute and stale chips gone as categories and every miner holding and following the chain. THE RESEARCH LANE's READING of the three shapes (21:4x UK; its identity edge = (E_card + F) / (E_mem + k F), F the GPU's premium per hash, k the chip core's energy per op over the GPU's): at the 1,400 MHz lock (v3 1.69 uJ per hash, v4 2.34, the premium 0.65 uJ = 6.5 pJ per counted op) the premium-free edge is 3.6x on GDDR7 and 5.3x on one HBM3 stack, 2.1x at F = 0.65 and k = 1, the asymptote 1/k; the premium needed for 2x at k = 1 is 0.76 uJ (103 W) at the lock and 1.41 uJ (175 W) at the stock point; at k at or under 0.5 no premium reaches 2x; a premium of zero is impossible by any hash-side lever (a joule the chip must spend is a joule the GPU spends first). Shape 1 (the shadow at the floor clock): correct and measured by the 1,400 row; it halves the premium and leaves k as the whole question (the edge's derivative in F at F = 0 is minus 5.7x per uJ at k = 1 and minus 0.2x per uJ at k = 0.3; the family mix with the highest k, shuffles and multiplies, is the right content; the tensor block has k near 1 but 0.056 pJ per MAC, so it forces no joules without 26x the verifier). Shape 2 (per-read work a chip cannot amortise): no construction found (the chip's lane count is set by its own latency, lane state is SRAM at pJ per access, extra reads scale both rates; row, bank, refresh and burst shaping have no asymmetry); one candidate for a measurement, not a claim: a hot table kept L2-resident through cache-policy hints (dataset loads evict-first, hot loads evict-last), since a 64 MiB SRAM read on a chip is not cheaper than a GPU L2 hit (the 3.0 layer 5 measurement used no hints and lost 13 to 16 percent). Shape 3 (the refresh as the cost): dead by arithmetic (a 1 GiB rebuild about 0.1 J on a chip against 280 kJ of hashing an hour). What lowers the premium-free floor is the card's own E_card: the 5090's idle 74 to 91 W plus its memory system's 55 W bound the floor near 2.0x on GDDR7 at any operating point; the two measurements that read how much of the 100 W between is reachable are the knee below 1,400 MHz (the floor pass, on PC 1's queue) and an SM-sparse kernel (the hash on a fraction of the SMs with 4 to 8 chains per thread, the rest clock-gated), both ordered to the hash lane with the L2 cache-policy hot table as PC 1 queue tail items (worker launch shapes, --bench only, no consensus change). THE FIVE RATIOS (adv-accept, 21:35 BST, at the acceptance's 2^20 sample, closed form and live to 1e-4): 4346 (bbb38e847011c354) minimum site 2 at 0.9840 REFUSED by (c''') at 0.995; 5245 (beaad44840bb9e4e) site 8 at 0.9831 REFUSED; 106924 at 0.9821 and 107022 at 0.9877 REFUSED; Devnet 3's epoch-0 class v4 program (fce15bf61030be57) site 0 at 0.9992 (live 0.9993) MISSED; era-fixed-20 (11f9f955b21d56c9) site 11 at 0.9997 MISSED; era-drawn-2 (7ceb797d31eedb3e) site 13 at 0.9992 MISSED. THE DEFENDER'S RULING: the (c''') floor at 0.995 closes the HOT-SET HALF of the class (every program with X_f at or above f that any lane measured, 100767 included) and NOT the class; the residual is the shadow-block-written value-level concentration below a ratio floor's resolution (1.26x to 1.45x of the top 0.1 percent share, 0.03 to 0.1 percent of a hash's reads each, minimum sites 0.9992 to 0.9997, which no floor reaches without sitting inside the clean seeds' own spread), chip gain under 1.001x, its lever a value-level source test at live scale or a per-site hot-item test, routed to the next class as a named item; section 14 of the v5 design says so with the five numbers verbatim; the freeze proceeds on the suite's green, both the floor's number and its reach measured. adv-mixer-2 (49656c2e) confirms AP-F4-1's class from outsider inputs (model A an FPGA LUT-area gain for a per-day build, 1.0 on every GPU, verifier and chip with a general multiplier; P(A at or above 1.1x) = 2^-10.8 per day on the exact median 226; model C under 2^-20 at 1.1x), the two censuses' medians (226 against 231) to be reconciled side by side in the record; its redraw rule to the v5 lane as a second independent statement. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh run by path under the inline-rm rule (v5-kits 4f9d96d2); the kit zip stands. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From 95b0aae3e57ef69e832222578bf16e3719e86079 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:43:53 +0000 Subject: [PATCH 07/15] Public chip text: the weak-day tail reads 15 days a century at the reconciled median 226 (litepaper table, /claims, evidence row 17); Counter ASIC 3.0 status: main's rulings (the freeze proceeds with the residual named; the halving ruled to the fleet), adv-cache-3 complete, the last-resort correctness finding Co-Authored-By: Claude Fable 5.1 --- docs/evidence.md | 2 +- docs/plans/counter-asic-3-public-text-2026-10-07.md | 4 ++-- docs/plans/counter-asic-3-status.md | 2 +- site/evidence.html | 2 +- site/litepaper.html | 2 +- 5 files changed, 6 insertions(+), 6 deletions(-) diff --git a/docs/evidence.md b/docs/evidence.md index 5f454be65..9cb19a950 100644 --- a/docs/evidence.md +++ b/docs/evidence.md @@ -41,7 +41,7 @@ Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml` | 14 | Ethereum bytecode runs unchanged, with the documented differences of spec 7.1 | Homepage Build card; litepaper Building | tested by the team | as row 13; fixes `F-exec-A`, `F-exec-B` (spec 7.5) | `tools/evm-smoke/smoke.mjs`: deploy via viem, `increment`, `hashLoop`, `eth_estimateGas`, `eth_getLogs`; `tools/exec-attacks` scenarios 1 and 3; bench-log "execution layer attack fixes" | Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463; the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The `Prover` precompile, proof records and the shard planner are not in the node | none yet | | 15 | Every block is proven, with the proof landing within about a minute at launch | Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate | implemented | repo `d7e1f89` (GPU proof), `e01a3cc`, `292e800`, `eedd136` (`proving/igneum-prove`: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6 | `proving/windows-wsl2` (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; `igneum-prove-host --mode block` on `proving/fixtures/`; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards" | First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture `block-78-increment` (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in `docs/benchmarks/proving-e2e.md`. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on PC 2 in 34 s, verified on the Mac in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind `proving_v1_activation_daa` (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is one | none yet | | 16 | A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves) | Litepaper Proving ("The proving budget"); roadmap gate 2 | designed | spec 5.1 (Target), 7.6 (`S_p` provisional, 7,500,000 pgas = `B_p` / 4) | `PROVE-SHARD.bat` on the RTX 5090 (pending); the end-to-end standard in `docs/benchmarks/proving-e2e.md`; bench-log "proving: devnet v4 shards" | Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional `S_p` is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB card | none yet | -| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 12 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 12 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026). | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | +| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026). | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | | 18 | The chip resistance measurements: the program is latency-bound (random reads), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache | Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page | tested by the team | readwidth e752fc7 (`docs/plans/read-width.md`), ca2-era 78c0ee4, ca2-cache 2de19e5 (`docs/plans/hot-table.md`) | The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per load | Latency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026 | none yet | | 19 | The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors | Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page | tested by the team | ca2-mixer 1ab8b21 (`tests/mixer.rs`, `tests/scratch.rs`), ca2-era 78c0ee4, ca2-soundness a465881 (`docs/analysis/scratch-soundness.md`), `igneum-pow/tests/packs.rs` | The crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per card | Class v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (PC 1 job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing) | none yet | | 20 | No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%) | Homepage stats and Economics tiles; litepaper Supply, Economics | implemented | repo `6ac80a3`; fork "igneum-node devnet v0"; `consensus/core/src/igneum.rs`, `coinbase.rs` | `cargo test -p kaspa-consensus-core igneum` (8 pass: subsidy table, ramp, split, cap) and `cargo test -p kaspa-consensus coinbase` (8 pass); `igneum-miner inspect 40`; bench-log "igneum-node devnet v0" | Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the `igneum-proving-pool-v0` output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happened | none yet | diff --git a/docs/plans/counter-asic-3-public-text-2026-10-07.md b/docs/plans/counter-asic-3-public-text-2026-10-07.md index a5f36ecb2..395f618f1 100644 --- a/docs/plans/counter-asic-3-public-text-2026-10-07.md +++ b/docs/plans/counter-asic-3-public-text-2026-10-07.md @@ -16,7 +16,7 @@ The chip model. We price the strongest chip we can design against an RTX 5090 an | The same chip at the ladder's second rung (about 200,000 ops per hash), reached by miner signal | about 2.8x | modelled, 7 October 2026 | | Any chip under class v5, where the dataset is the chain's own state | a stateless or stale chip is wrong on every item, so the stored-dataset chip and the recompute chip are removed as categories; the verifier pays 0.2 ms more per warp | designed, 7 October 2026 | | A chip caching the hottest 0.1 percent of items (about 1 MB of SRAM) | bounded at 1.067x at the ceiling, 1.005x on about half the hours and 1.048x on 5 percent | measured census of 1,024 programs, 7 October 2026; the source rule in the next class | -| A per-day FPGA that recomputes the dataset with cheap multipliers on a weak day | at most 12 percent more hash rate on 12 days a century, nothing on the other days and nothing for any chip | measured census of 2^24 days, 7 October 2026; the rule in the next class | +| A per-day FPGA that recomputes the dataset with cheap multipliers on a weak day | at most 12 percent more multiplier area on an FPGA's per-day build on 15 days a century, nothing on the other days and nothing for any chip | measured census of 2^24 days, 7 October 2026; the rule in the next class | | When a stored-dataset chip pays for itself | at about USD 100 M of market cap in the first two years, not before | modelled, 7 October 2026 | | The baseline the work started from: the same chip under class v3, without the shadow (the Ethash class) | 5x to 9x (5.1x on GDDR7, 9.2x on eight HBM3 stacks; the Ethash chips of this class reached 2.1x to 4.8x) | modelled, 6 October 2026; the precedent measured by others, 2020 to 2022; never the launch state | @@ -30,4 +30,4 @@ Your card against the strongest chip we can price: an RTX 5090 at 136 MH/s on 35 | # | Claim | Where it is made | Status | Version or commit | Reproducible test | Result, date, machine | Independent verification | |---|---|---|---|---|---|---|---| -| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 12 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 12 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | +| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index 344e5ca6b..5d6397dcb 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -455,7 +455,7 @@ Reading so far: the class v4 premium at the unlocked clock is +145 W on this car | 1,400 | 134.98 | 316.3 | 0.427 | 134.68 | 228.0 | 0.591 | 1,387 | | unlocked, end | 136.75 | 473.0 | | 136.5x | 329.7 | | 2,843 / 2,850 | -Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). THE FOUNDER'S QUESTION ON THE CLASS V4 PREMIUM ("we need a solution, deep research, other methods, something must be doable even if it is revolutionary"; a research lane, counter-asic-4, on the literature and the alternatives by 23:30Z). THE ONE COMPUTATION FROM THIS LANE (chip-model-v3 section 5.10, on the mirror's master at aa829826): class v5 ON and the shadow at ZERO leaves the strongest chip, the f = 1 stored-dataset chip, at 5.1x (GDDR7, the 5090's own 16 devices without the GPU: 166 MH/s at 78 W with a farm-shared node) to 9.1x (HBM3 eight stacks: 666 MH/s at 175 W) per joule over the 5090's 0.417 MH/W, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost (class-v5 2a.2: one node serves a farm, the leaves ship at 16.5 KB/s to 10,000 members, the rebuild is the same 32 ms per window every GPU pays); only a chip forced to carry its own node (85 W, approximate) falls near 2x, and only the small ones (one HBM3 stack 1.8x, the GDDR7 board 2.5x; the eight-stack package 6.2x). So the shadow stays the only lever in the model that reaches the memory-system chip, and the premium is its price; what class v5 buys is the recompute and stale chips gone as categories and every miner holding and following the chain. THE RESEARCH LANE's READING of the three shapes (21:4x UK; its identity edge = (E_card + F) / (E_mem + k F), F the GPU's premium per hash, k the chip core's energy per op over the GPU's): at the 1,400 MHz lock (v3 1.69 uJ per hash, v4 2.34, the premium 0.65 uJ = 6.5 pJ per counted op) the premium-free edge is 3.6x on GDDR7 and 5.3x on one HBM3 stack, 2.1x at F = 0.65 and k = 1, the asymptote 1/k; the premium needed for 2x at k = 1 is 0.76 uJ (103 W) at the lock and 1.41 uJ (175 W) at the stock point; at k at or under 0.5 no premium reaches 2x; a premium of zero is impossible by any hash-side lever (a joule the chip must spend is a joule the GPU spends first). Shape 1 (the shadow at the floor clock): correct and measured by the 1,400 row; it halves the premium and leaves k as the whole question (the edge's derivative in F at F = 0 is minus 5.7x per uJ at k = 1 and minus 0.2x per uJ at k = 0.3; the family mix with the highest k, shuffles and multiplies, is the right content; the tensor block has k near 1 but 0.056 pJ per MAC, so it forces no joules without 26x the verifier). Shape 2 (per-read work a chip cannot amortise): no construction found (the chip's lane count is set by its own latency, lane state is SRAM at pJ per access, extra reads scale both rates; row, bank, refresh and burst shaping have no asymmetry); one candidate for a measurement, not a claim: a hot table kept L2-resident through cache-policy hints (dataset loads evict-first, hot loads evict-last), since a 64 MiB SRAM read on a chip is not cheaper than a GPU L2 hit (the 3.0 layer 5 measurement used no hints and lost 13 to 16 percent). Shape 3 (the refresh as the cost): dead by arithmetic (a 1 GiB rebuild about 0.1 J on a chip against 280 kJ of hashing an hour). What lowers the premium-free floor is the card's own E_card: the 5090's idle 74 to 91 W plus its memory system's 55 W bound the floor near 2.0x on GDDR7 at any operating point; the two measurements that read how much of the 100 W between is reachable are the knee below 1,400 MHz (the floor pass, on PC 1's queue) and an SM-sparse kernel (the hash on a fraction of the SMs with 4 to 8 chains per thread, the rest clock-gated), both ordered to the hash lane with the L2 cache-policy hot table as PC 1 queue tail items (worker launch shapes, --bench only, no consensus change). THE FIVE RATIOS (adv-accept, 21:35 BST, at the acceptance's 2^20 sample, closed form and live to 1e-4): 4346 (bbb38e847011c354) minimum site 2 at 0.9840 REFUSED by (c''') at 0.995; 5245 (beaad44840bb9e4e) site 8 at 0.9831 REFUSED; 106924 at 0.9821 and 107022 at 0.9877 REFUSED; Devnet 3's epoch-0 class v4 program (fce15bf61030be57) site 0 at 0.9992 (live 0.9993) MISSED; era-fixed-20 (11f9f955b21d56c9) site 11 at 0.9997 MISSED; era-drawn-2 (7ceb797d31eedb3e) site 13 at 0.9992 MISSED. THE DEFENDER'S RULING: the (c''') floor at 0.995 closes the HOT-SET HALF of the class (every program with X_f at or above f that any lane measured, 100767 included) and NOT the class; the residual is the shadow-block-written value-level concentration below a ratio floor's resolution (1.26x to 1.45x of the top 0.1 percent share, 0.03 to 0.1 percent of a hash's reads each, minimum sites 0.9992 to 0.9997, which no floor reaches without sitting inside the clean seeds' own spread), chip gain under 1.001x, its lever a value-level source test at live scale or a per-site hot-item test, routed to the next class as a named item; section 14 of the v5 design says so with the five numbers verbatim; the freeze proceeds on the suite's green, both the floor's number and its reach measured. adv-mixer-2 (49656c2e) confirms AP-F4-1's class from outsider inputs (model A an FPGA LUT-area gain for a per-day build, 1.0 on every GPU, verifier and chip with a general multiplier; P(A at or above 1.1x) = 2^-10.8 per day on the exact median 226; model C under 2^-20 at 1.1x), the two censuses' medians (226 against 231) to be reconciled side by side in the record; its redraw rule to the v5 lane as a second independent statement. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh run by path under the inline-rm rule (v5-kits 4f9d96d2); the kit zip stands. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). THE FOUNDER'S QUESTION ON THE CLASS V4 PREMIUM ("we need a solution, deep research, other methods, something must be doable even if it is revolutionary"; a research lane, counter-asic-4, on the literature and the alternatives by 23:30Z). THE ONE COMPUTATION FROM THIS LANE (chip-model-v3 section 5.10, on the mirror's master at aa829826): class v5 ON and the shadow at ZERO leaves the strongest chip, the f = 1 stored-dataset chip, at 5.1x (GDDR7, the 5090's own 16 devices without the GPU: 166 MH/s at 78 W with a farm-shared node) to 9.1x (HBM3 eight stacks: 666 MH/s at 175 W) per joule over the 5090's 0.417 MH/W, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost (class-v5 2a.2: one node serves a farm, the leaves ship at 16.5 KB/s to 10,000 members, the rebuild is the same 32 ms per window every GPU pays); only a chip forced to carry its own node (85 W, approximate) falls near 2x, and only the small ones (one HBM3 stack 1.8x, the GDDR7 board 2.5x; the eight-stack package 6.2x). So the shadow stays the only lever in the model that reaches the memory-system chip, and the premium is its price; what class v5 buys is the recompute and stale chips gone as categories and every miner holding and following the chain. THE RESEARCH LANE's READING of the three shapes (21:4x UK; its identity edge = (E_card + F) / (E_mem + k F), F the GPU's premium per hash, k the chip core's energy per op over the GPU's): at the 1,400 MHz lock (v3 1.69 uJ per hash, v4 2.34, the premium 0.65 uJ = 6.5 pJ per counted op) the premium-free edge is 3.6x on GDDR7 and 5.3x on one HBM3 stack, 2.1x at F = 0.65 and k = 1, the asymptote 1/k; the premium needed for 2x at k = 1 is 0.76 uJ (103 W) at the lock and 1.41 uJ (175 W) at the stock point; at k at or under 0.5 no premium reaches 2x; a premium of zero is impossible by any hash-side lever (a joule the chip must spend is a joule the GPU spends first). Shape 1 (the shadow at the floor clock): correct and measured by the 1,400 row; it halves the premium and leaves k as the whole question (the edge's derivative in F at F = 0 is minus 5.7x per uJ at k = 1 and minus 0.2x per uJ at k = 0.3; the family mix with the highest k, shuffles and multiplies, is the right content; the tensor block has k near 1 but 0.056 pJ per MAC, so it forces no joules without 26x the verifier). Shape 2 (per-read work a chip cannot amortise): no construction found (the chip's lane count is set by its own latency, lane state is SRAM at pJ per access, extra reads scale both rates; row, bank, refresh and burst shaping have no asymmetry); one candidate for a measurement, not a claim: a hot table kept L2-resident through cache-policy hints (dataset loads evict-first, hot loads evict-last), since a 64 MiB SRAM read on a chip is not cheaper than a GPU L2 hit (the 3.0 layer 5 measurement used no hints and lost 13 to 16 percent). Shape 3 (the refresh as the cost): dead by arithmetic (a 1 GiB rebuild about 0.1 J on a chip against 280 kJ of hashing an hour). What lowers the premium-free floor is the card's own E_card: the 5090's idle 74 to 91 W plus its memory system's 55 W bound the floor near 2.0x on GDDR7 at any operating point; the two measurements that read how much of the 100 W between is reachable are the knee below 1,400 MHz (the floor pass, on PC 1's queue) and an SM-sparse kernel (the hash on a fraction of the SMs with 4 to 8 chains per thread, the rest clock-gated), both ordered to the hash lane with the L2 cache-policy hot table as PC 1 queue tail items (worker launch shapes, --bench only, no consensus change). THE FIVE RATIOS (adv-accept, 21:35 BST, at the acceptance's 2^20 sample, closed form and live to 1e-4): 4346 (bbb38e847011c354) minimum site 2 at 0.9840 REFUSED by (c''') at 0.995; 5245 (beaad44840bb9e4e) site 8 at 0.9831 REFUSED; 106924 at 0.9821 and 107022 at 0.9877 REFUSED; Devnet 3's epoch-0 class v4 program (fce15bf61030be57) site 0 at 0.9992 (live 0.9993) MISSED; era-fixed-20 (11f9f955b21d56c9) site 11 at 0.9997 MISSED; era-drawn-2 (7ceb797d31eedb3e) site 13 at 0.9992 MISSED. THE DEFENDER'S RULING: the (c''') floor at 0.995 closes the HOT-SET HALF of the class (every program with X_f at or above f that any lane measured, 100767 included) and NOT the class; the residual is the shadow-block-written value-level concentration below a ratio floor's resolution (1.26x to 1.45x of the top 0.1 percent share, 0.03 to 0.1 percent of a hash's reads each, minimum sites 0.9992 to 0.9997, which no floor reaches without sitting inside the clean seeds' own spread), chip gain under 1.001x, its lever a value-level source test at live scale or a per-site hot-item test, routed to the next class as a named item; section 14 of the v5 design says so with the five numbers verbatim; the freeze proceeds on the suite's green, both the floor's number and its reach measured. adv-mixer-2 (49656c2e) confirms AP-F4-1's class from outsider inputs (model A an FPGA LUT-area gain for a per-day build, 1.0 on every GPU, verifier and chip with a general multiplier; P(A at or above 1.1x) = 2^-10.8 per day on the exact median 226; model C under 2^-20 at 1.1x), the two censuses' medians (226 against 231) to be reconciled side by side in the record; its redraw rule to the v5 lane as a second independent statement. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh run by path under the inline-rm rule (v5-kits 4f9d96d2); the kit zip stands. MAIN'S RULINGS (21:4x UK). (1) The class v5 freeze does not wait on the three milder concentrations the 0.995 floor misses (Devnet 3's first program among them): their chip gain is bounded under 1.002x by their size, so the freeze proceeds with the floor as it is, the record names them as the residual, and the fix question (a value-level source test at live scale) is a 0.3.25 item. (2) The weak-day census reconciled at median 226 (F4's NAF weight had counted the carry digit at position 32, which a 32-bit multiplier never pays; 0.333 digits per word, 5.3 adders per day, the whole of 231 against 226 and of 12 against 15 days a century, the same worst day in both; F4's record section 9 carries both medians side by side, mirrored in adv-mixer-2's report): 5.69e-4 of days (2^-10.8), 15 days a century over 1.1x on FPGA LUT area, worst 2050-04-28 at 1.113x, the DSP-bound readings at 0 for k at least 2, F4's PASS against class v4 unchanged; every public text saying "12 days a century" becomes "15 days a century" (done in this landing on the litepaper table, /claims through it, evidence row 17 and the public text file, the sentence now "at most 12 percent more multiplier area on an FPGA's per-day build on 15 days a century, nothing on the other days and nothing for any chip"); the site audit lane told. (3) THE HALVING, ruled to the fleet (21:4x UK), neither (a) nor (b) as written: on every standing box the OLD shared-devnet miner stops and the Devnet 3 miner keeps the card at full rate (one miner per card, the exception over); hub-1, pool-1 and the live seed stay on the old chain as its voters and miners until every poller has moved; the old-chain nodes on the boxes stay up as voters without mining; no 3070 fallback; the fleet reports per-box restore lines. THE IN-HOUSE PASS, MORE CLOSES: adv-cache-3 COMPLETE (bc2d01d5, 0.23 slot-hours; every row BOUND or PASS, every plant fired; the exhaustive w = 2 image census at all 64 depths equal to the random-function recursion to 2 x 10^-5; the flip-table ladder at 2^18 lines no single-bit bias from two double rounds up). adv-accept-3 (aa359962): a correctness FINDING for the rule's owners, no chain consequence: the deterministic last-resort program (after 256 failed attempts) FAILS the real rule in 223 of 2,500 seeds (209 by part (a), a cyclic stale load) yet is handed to the chain unchecked, unreachable at 4.6e-44 per epoch; ruled to class v5's generator as the commit after the freeze (the last resort passing the rule by construction, known-failed first on one of the 223 seeds; sub-version 3's stated as unreachable and unverified, class v5's as verified); also the 256-unit stand-in ratio is noise as a selector (the three lowest of 4,975 clean live at 1.0002x worst), the selector working only at the 2^20-unit read. Lanes complete or at a natural end: adv-cache, adv-accept-2, adv-cache-3; adv-mixer on its last solve. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh by path (v5-kits 4f9d96d2) under the inline-rm rule. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule diff --git a/site/evidence.html b/site/evidence.html index c92824678..a7fb3b72b 100644 --- a/site/evidence.html +++ b/site/evidence.html @@ -254,7 +254,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v 14Ethereum bytecode runs unchanged, with the documented differences of spec 7.1
    Homepage Build card; litepaper Building
    tested by the teamas row 13; fixes F-exec-A, F-exec-B (spec 7.5)tools/evm-smoke/smoke.mjs: deploy via viem, increment, hashLoop, eth_estimateGas, eth_getLogs; tools/exec-attacks scenarios 1 and 3; bench-log "execution layer attack fixes"Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463; the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The Prover precompile, proof records and the shard planner are not in the nodenone yet 15Every block is proven, with the proof landing within about a minute at launch
    Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate
    implementedrepo d7e1f89 (GPU proof), e01a3cc, 292e800, eedd136 (proving/igneum-prove: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6proving/windows-wsl2 (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; igneum-prove-host --mode block on proving/fixtures/; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards"First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture block-78-increment (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in docs/benchmarks/proving-e2e.md. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on the RTX 5090 Windows rig in 34 s, verified on the Apple M5 Max in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind proving_v1_activation_daa (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is onenone yet 16A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves)
    Litepaper Proving ("The proving budget"); roadmap gate 2
    designedspec 5.1 (Target), 7.6 (S_p provisional, 7,500,000 pgas = B_p / 4)PROVE-SHARD.bat on the RTX 5090 (pending); the end-to-end standard in docs/benchmarks/proving-e2e.md; bench-log "proving: devnet v4 shards"Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional S_p is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB cardnone yet -17The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 12 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state)
    the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
    tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measureddocs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/plans/counter-asic-3-status.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; docs/design/class-v5-stored-state.md; the H100 and market-cap rows of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 12 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, the RTX 5090 Windows rig's RTX 5090, the three-card Windows rig's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026).none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word +17The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state)
    the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
    tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measureddocs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/plans/counter-asic-3-status.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; docs/design/class-v5-stored-state.md; the H100 and market-cap rows of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, the RTX 5090 Windows rig's RTX 5090, the three-card Windows rig's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026).none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word 18The chip resistance measurements: the program is latency-bound (random reads), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache
    Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page
    tested by the teamreadwidth e752fc7 (docs/plans/read-width.md), ca2-era 78c0ee4, ca2-cache 2de19e5 (docs/plans/hot-table.md)The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per loadLatency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026none yet 19The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors
    Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page
    tested by the teamca2-mixer 1ab8b21 (tests/mixer.rs, tests/scratch.rs), ca2-era 78c0ee4, ca2-soundness a465881 (docs/analysis/scratch-soundness.md), igneum-pow/tests/packs.rsThe crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per cardClass v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (the three-card Windows rig (RTX 5090, RTX 4070, RX 9070 XT) job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing)none yet 20No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%)
    Homepage stats and Economics tiles; litepaper Supply, Economics
    implementedrepo 6ac80a3; fork "igneum-node devnet v0"; consensus/core/src/igneum.rs, coinbase.rscargo test -p kaspa-consensus-core igneum (8 pass: subsidy table, ramp, split, cap) and cargo test -p kaspa-consensus coinbase (8 pass); igneum-miner inspect 40; bench-log "igneum-node devnet v0"Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the igneum-proving-pool-v0 output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happenednone yet diff --git a/site/litepaper.html b/site/litepaper.html index b2d0903a9..8a9b13fcc 100644 --- a/site/litepaper.html +++ b/site/litepaper.html @@ -445,7 +445,7 @@ body.all .pager{display:none} The same chip at the ladder’s second rung (about 200,000 ops per hash), reached by miner signalabout 2.8xmodelled, 7 October 2026 Any chip under class v5, where the dataset is the chain’s own statea stateless or stale chip is wrong on every item, so the stored-dataset chip and the recompute chip are removed as categories; the verifier pays 0.2 ms more per warpdesigned, 7 October 2026 A chip caching the hottest 0.1 percent of items (about 1 MB of SRAM)bounded at 1.067x at the ceiling, 1.005x on about half the hours and 1.048x on 5 percentmeasured census of 1,024 programs, 7 October 2026; the source rule in the next class - A per-day FPGA that recomputes the dataset with cheap multipliers on a weak dayat most 12 percent more hash rate on 12 days a century, nothing on the other days and nothing for any chipmeasured census of 2^24 days, 7 October 2026; the rule in the next class + A per-day FPGA that recomputes the dataset with cheap multipliers on a weak dayat most 12 percent more multiplier area on an FPGA’s per-day build on 15 days a century, nothing on the other days and nothing for any chipmeasured census of 2^24 days, 7 October 2026; the rule in the next class When a stored-dataset chip pays for itselfat about USD 100 M of market cap in the first two years, not beforemodelled, 7 October 2026 The baseline the work started from: the same chip under class v3, without the shadow (the Ethash class)5x to 9x (5.1x on GDDR7, 9.2x on eight HBM3 stacks; the Ethash chips of this class reached 2.1x to 4.8x)modelled, 6 October 2026; the precedent measured by others, 2020 to 2022; never the launch state
    From 3f5a642c79428d8b04ebefd685b0022fdf1265bf Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:46:51 +0000 Subject: [PATCH 08/15] release rules 5 extended (digest-moving node releases move every node inside one minute, the apps' entries after); release 0.3.22 section 12: the 2720d8d2 split, the revert, the pulled-back entries, the pool split height, PC 1's helper hang Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 4 ++++ docs/plans/release-rules.md | 2 +- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index 03ef2b509..d59906804 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -122,3 +122,7 @@ Three commits on 96161037: bd710a36 the sub-version 3 re-pin (byte 7, igneum-pow **LG-4 (the founder: "use PC 1 or PC 2 or get another machine"; main's ruling):** on PC 2 tonight after both smokes, a fresh Windows user account created by job (no Igneum state, no card cache), the three timed steps (download and install, sync to the tip, dataset build to the first accepted share) as FIRST-SHARE lines, ten runs with the app uninstalled and the profile wiped between runs, no click anywhere, the rows on /evidence labelled "PC 2, fresh user account, not a fresh image, 7 October 2026", the 9-of-10 under-10-minutes bar read against them; macOS the same way as a fresh user account on the Mac tomorrow; a rented Windows VM only if the account shape fails the bar for a reason an image would change. PC 1 stays the founder's desk. **0.3.22 Windows take 2 FAIL (21:13 BST) and the skip:** the job's detached helper was ended with the job's process tree before its first sleep ran out (Start-Process stays in the runner's tree; a survivor needs Win32_Process.Create or a scheduled task); nothing was installed, every file still 0.3.21, the payload untouched. Ruling: no take 3; the 0.3.22 Windows entry is skipped (the Mac and HiveOS entries stand); the first install over a running app is 0.3.23's installer with install-close-23 in the simplest job shape (Start-Process -Wait; the installer's own stop step and the install-running flag do the work), its smoke read the gate line for both; the Discord card publishes on the 0.3.23 Windows entry. **The version miss, second layer (21:30 BST):** the box cross of 0.3.23's exes failed in build.rs because igneum-app.rc still read 0.3.22 (Info.plist and the installer's AppVersion too); fixed at release-0.3.23 = 7c7489ac, the rule 15 check extended to six places and green on the tree; the 0.3.23 node pairs under /srv/artefacts/0323-2720d8d2/ (hands f2cf6a87/fb147dd1, seed 3edaf83d/be5ca735, win 8326d78a/38c74545) with the engine string. + +## 12. The 2720d8d2 split and the one-minute form (21:1x to 21:4x BST) + +The 0.3.23 node's heights move Devnet 3's digest to ba75bf6f, and the one-box-at-a-time sweep to it split the network: dn3-g1 flipped at 21:18 BST and sat alone twenty minutes on its own fork (blocks 12,553 to 12,793, peers 0; the 83eb50cd nodes refuse ba75bf6f at the handshake; build-1's seed logged 27 digest-mismatch rejects by 21:37); the fleet stopped the pass at 21:31 after that one box (the Vast ssh proxies ssh1, 3, 4, 5, 6 refused the next five, so nothing else flipped) and reverted dn3-g1 to 34a2dbaa on its kept datadir, where it reorgs onto the network's heavier chain; no record or share of the network's sat on the solo blocks; the pool pair, the provers and the second nodes stayed on 83eb50cd. The shipper's part of the fault: the 0.3.23 Mac entry (both folders, 21:39:09 BST, DMG 4ada2807 on 7c7489ac) and the 0.3.23 HiveOS alias (0d716ed9, 21:37) were published ahead of the fleet's move, against rule 5's own word; both pulled back to 0.3.22 (live again at 21:41:08 and 21:44:48 BST; the Mac still runs 0.3.22; the 0.3.23 hive package held in scratch r0323/held-public). The move to 2720d8d2 is the 0.3.20 form (rule 5 extended): binaries pre-placed on every Devnet 3 node (the fleet's thirty, build-1's three, the pool pair's two), every node restarted inside one minute at a clock the fleet names ten minutes ahead once the proxies answer, the digest read back per node, the first lock on the new side as the line; the 0.3.23 Mac entry, the hive alias and the Windows entry publish at that minute or after. Pool split: the hour's 13,209 + 7,200 = 20,409 lands before the 0.3.24 publish minute, so the split goes into the 0.3.24 object at or after the v5 floor on IGNH/IGNW (pending main); the node lane holds it out of the v5 object commit. PC 1: the 5090 floor grid exited 0 at 21:41 BST but the Power Helper hung at its 1,000 MHz step and the 5090 mines at the 1,100 MHz lock until the hash lane's unelevated restore job runs; the grid gets the fix (stop at the first helper timeout, restart the task before the reset). diff --git a/docs/plans/release-rules.md b/docs/plans/release-rules.md index ecddb7a3a..3a4b8f5ab 100644 --- a/docs/plans/release-rules.md +++ b/docs/plans/release-rules.md @@ -9,7 +9,7 @@ Every cut of the Igneum Miner app and its node runs under these. The dated plan 4c. **The proving ids gate (main, 7 October 2026, after the 0.3.20 blocker).** On every candidate, a node started on the LIVE override file reports both proving ids (the shard program id and the aggregator id) on its proving v1 start line, and a prover's first statement against it is accepted; a zero-id statement is the known-failed shape. It runs beside the kept-datadir read, since both share the warm pod. Why: c4459193 read the ids as unknown, built statements with zeros and refused every proof; a sweep would have stopped every prover's pay. 4a. **Every gate starts on every candidate the moment its binary builds, never after the pin (the founder, 7 October 2026).** The digest and mixed-version gates, the kept-datadir start, the relay and poison cases and the wipe canary all begin on each candidate binary as it lands; a struck candidate's runs are stopped and its successor's begin. The post-pin wait is then the longest single form (about 80 minutes, the wipe), not the sum. 4b. **Warm pods per gate class (the founder, 7 October 2026, ordered to the fleet lane).** The fleet keeps synced pods warm for each gate class so a case form's target starts at the tip (a kept copy of the live line, caught up), never from a kept copy far behind it; the wipe canary is the only full IBD in the set. -5. **Rollout in waves, each box read back (the founder, 7 October 2026, replacing one-box-at-a-time):** PC 1 first, then PC 2, the Mac, the seed, the hands and the fleet in parallel waves as the lock lines allow; a lock line from the hub between waves; hold if the frozen table's signed share reads under 75; every box read back by its commit string. When the publish moves the consensus floor (a new digest), every 0.3.x node on the old file refuses the new ones as peers until it is swept, so the seed, the hands and the fleet move in the first wave with the apps' pollers, not last. Every lock line of a sweep that replaces nodes carrying a consensus floor names the date the sweep must finish (0.3.20: before 13 October 2026 09:00 UK). The wave list is written, not remembered: every sweep's first wave names each Hetzner seed by address (188.245.5.161:26611 for the shared devnet; the testnet seeds when they move) beside the hands and the fleet, and the seed's row closes only on its own read-back line (string, digest, first accepted block) from the lane that holds its key (the build-server lane, infra/devnet/restart-seed.sh). Added 7 October 2026 after the 0.3.20 sweep left the seed on the old object for one hour forty, found by the 0.3.21 wipe canary's reject lines. +5. **Rollout in waves, each box read back (the founder, 7 October 2026, replacing one-box-at-a-time):** PC 1 first, then PC 2, the Mac, the seed, the hands and the fleet in parallel waves as the lock lines allow; a lock line from the hub between waves; hold if the frozen table's signed share reads under 75; every box read back by its commit string. When the publish moves the consensus floor (a new digest), every 0.3.x node on the old file refuses the new ones as peers until it is swept, so the seed, the hands and the fleet move in the first wave with the apps' pollers, not last. Every lock line of a sweep that replaces nodes carrying a consensus floor names the date the sweep must finish (0.3.20: before 13 October 2026 09:00 UK). The wave list is written, not remembered: every sweep's first wave names each Hetzner seed by address (188.245.5.161:26611 for the shared devnet; the testnet seeds when they move) beside the hands and the fleet, and the seed's row closes only on its own read-back line (string, digest, first accepted block) from the lane that holds its key (the build-server lane, infra/devnet/restart-seed.sh). Added 7 October 2026 after the 0.3.20 sweep left the seed on the old object for one hour forty, found by the 0.3.21 wipe canary's reject lines. **Digest-moving node releases (7 October 2026, after the 2720d8d2 split):** when the node commit moves the consensus params digest (a height, a floor, a new switch field), a one-box-at-a-time sweep is a network split (dn3-g1 sat alone twenty minutes on its own fork at 21:18 BST), so the move is the 0.3.20 floor-move form: the binaries pre-placed on every node of the network (the fleet's, the hands, the seeds, the pool pair, every second node), every node restarted inside one minute at a clock named ten minutes ahead, the digest read back per node after, the first lock on the new side as the line; the apps' entries (manifest, HiveOS alias) publish at that minute or after, never before, so no app sits alone on the new digest. The one-box form is for digest-preserving releases only (34a2dbaa tonight). The release note names which form a release takes before its first box moves. 6. **Read-back is by commit string plus digest plus engine:** on 0.3.18+ nodes igneum_getNodeInfo powEngine must read "igneum-pow" ("stub" = FAIL); on earlier trees `strings igneumd | grep -c igneum-pow/src/` above zero. The miner embeds no commit string; its pairing is the build line and the sha. 7. **igneum-pow pairing:** a fork build takes igneum-pow by path from the igneum worktree it sits in; build each node tree inside its own app worktree whose igneum-pow is the pinned tree; the pairing log line names it. Master's build tools need rust-toolchain.toml in the tree (the app tree's pin applies to a vendor worktree under it; a standalone node checkout is unpinned until the node line carries its own file). Extended 7 October 2026 (the Devnet 3 pool pair, three WRONG HASH rounds each a tree a step behind the chain): the rule binds every crate that embeds kaspa-pow, the node, the pool daemon (igneum-pool) and the app's CPU re-check alike, each built against the pinned igneum-pow by path from the release worktree (017e7037 on 0.3.22 and 0.3.23), and the packs pin travels with the generator (the proto-cuda packs are the pin's export, never an older one); the read-back is the paired miner's "class v4 program id <16 hex>" line equal to the node's per epoch, and the daemon accepting its shares. 8. **glibc classes:** HiveOS 2.31 (`--ship hive`, smoke in ubuntu:20.04 on the box), seeds and generic 2.35 (`--ship seed`), fleet 24.04 boxes native 2.39. From 532d19853c4647c177f230bd9b32c0b27bf57b32 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:47:43 +0000 Subject: [PATCH 09/15] release rules 5: the publisher's digest gate check (an app entry whose pin's digest differs from the network's is refused without the move's clock) Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-rules.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/release-rules.md b/docs/plans/release-rules.md index 3a4b8f5ab..9f9282b2d 100644 --- a/docs/plans/release-rules.md +++ b/docs/plans/release-rules.md @@ -9,7 +9,7 @@ Every cut of the Igneum Miner app and its node runs under these. The dated plan 4c. **The proving ids gate (main, 7 October 2026, after the 0.3.20 blocker).** On every candidate, a node started on the LIVE override file reports both proving ids (the shard program id and the aggregator id) on its proving v1 start line, and a prover's first statement against it is accepted; a zero-id statement is the known-failed shape. It runs beside the kept-datadir read, since both share the warm pod. Why: c4459193 read the ids as unknown, built statements with zeros and refused every proof; a sweep would have stopped every prover's pay. 4a. **Every gate starts on every candidate the moment its binary builds, never after the pin (the founder, 7 October 2026).** The digest and mixed-version gates, the kept-datadir start, the relay and poison cases and the wipe canary all begin on each candidate binary as it lands; a struck candidate's runs are stopped and its successor's begin. The post-pin wait is then the longest single form (about 80 minutes, the wipe), not the sum. 4b. **Warm pods per gate class (the founder, 7 October 2026, ordered to the fleet lane).** The fleet keeps synced pods warm for each gate class so a case form's target starts at the tip (a kept copy of the live line, caught up), never from a kept copy far behind it; the wipe canary is the only full IBD in the set. -5. **Rollout in waves, each box read back (the founder, 7 October 2026, replacing one-box-at-a-time):** PC 1 first, then PC 2, the Mac, the seed, the hands and the fleet in parallel waves as the lock lines allow; a lock line from the hub between waves; hold if the frozen table's signed share reads under 75; every box read back by its commit string. When the publish moves the consensus floor (a new digest), every 0.3.x node on the old file refuses the new ones as peers until it is swept, so the seed, the hands and the fleet move in the first wave with the apps' pollers, not last. Every lock line of a sweep that replaces nodes carrying a consensus floor names the date the sweep must finish (0.3.20: before 13 October 2026 09:00 UK). The wave list is written, not remembered: every sweep's first wave names each Hetzner seed by address (188.245.5.161:26611 for the shared devnet; the testnet seeds when they move) beside the hands and the fleet, and the seed's row closes only on its own read-back line (string, digest, first accepted block) from the lane that holds its key (the build-server lane, infra/devnet/restart-seed.sh). Added 7 October 2026 after the 0.3.20 sweep left the seed on the old object for one hour forty, found by the 0.3.21 wipe canary's reject lines. **Digest-moving node releases (7 October 2026, after the 2720d8d2 split):** when the node commit moves the consensus params digest (a height, a floor, a new switch field), a one-box-at-a-time sweep is a network split (dn3-g1 sat alone twenty minutes on its own fork at 21:18 BST), so the move is the 0.3.20 floor-move form: the binaries pre-placed on every node of the network (the fleet's, the hands, the seeds, the pool pair, every second node), every node restarted inside one minute at a clock named ten minutes ahead, the digest read back per node after, the first lock on the new side as the line; the apps' entries (manifest, HiveOS alias) publish at that minute or after, never before, so no app sits alone on the new digest. The one-box form is for digest-preserving releases only (34a2dbaa tonight). The release note names which form a release takes before its first box moves. +5. **Rollout in waves, each box read back (the founder, 7 October 2026, replacing one-box-at-a-time):** PC 1 first, then PC 2, the Mac, the seed, the hands and the fleet in parallel waves as the lock lines allow; a lock line from the hub between waves; hold if the frozen table's signed share reads under 75; every box read back by its commit string. When the publish moves the consensus floor (a new digest), every 0.3.x node on the old file refuses the new ones as peers until it is swept, so the seed, the hands and the fleet move in the first wave with the apps' pollers, not last. Every lock line of a sweep that replaces nodes carrying a consensus floor names the date the sweep must finish (0.3.20: before 13 October 2026 09:00 UK). The wave list is written, not remembered: every sweep's first wave names each Hetzner seed by address (188.245.5.161:26611 for the shared devnet; the testnet seeds when they move) beside the hands and the fleet, and the seed's row closes only on its own read-back line (string, digest, first accepted block) from the lane that holds its key (the build-server lane, infra/devnet/restart-seed.sh). Added 7 October 2026 after the 0.3.20 sweep left the seed on the old object for one hour forty, found by the 0.3.21 wipe canary's reject lines. **Digest-moving node releases (7 October 2026, after the 2720d8d2 split):** when the node commit moves the consensus params digest (a height, a floor, a new switch field), a one-box-at-a-time sweep is a network split (dn3-g1 sat alone twenty minutes on its own fork at 21:18 BST), so the move is the 0.3.20 floor-move form: the binaries pre-placed on every node of the network (the fleet's, the hands, the seeds, the pool pair, every second node), every node restarted inside one minute at a clock named ten minutes ahead, the digest read back per node after, the first lock on the new side as the line; the apps' entries (manifest, HiveOS alias) publish at that minute or after, never before, so no app sits alone on the new digest. The one-box form is for digest-preserving releases only (34a2dbaa tonight). The release note names which form a release takes before its first box moves. Gate check (main, 7 October 2026): the publisher refuses an app entry whose node pin's digest (the pinned binary with no file on the entry's network) differs from the network's current digest (the hub's RPC) unless the move's clock is recorded in the entry's notes (--move-clock), known-failed self-test on tonight's shape; tooling on the 0.3.24 publish (the build-server lane). 6. **Read-back is by commit string plus digest plus engine:** on 0.3.18+ nodes igneum_getNodeInfo powEngine must read "igneum-pow" ("stub" = FAIL); on earlier trees `strings igneumd | grep -c igneum-pow/src/` above zero. The miner embeds no commit string; its pairing is the build line and the sha. 7. **igneum-pow pairing:** a fork build takes igneum-pow by path from the igneum worktree it sits in; build each node tree inside its own app worktree whose igneum-pow is the pinned tree; the pairing log line names it. Master's build tools need rust-toolchain.toml in the tree (the app tree's pin applies to a vendor worktree under it; a standalone node checkout is unpinned until the node line carries its own file). Extended 7 October 2026 (the Devnet 3 pool pair, three WRONG HASH rounds each a tree a step behind the chain): the rule binds every crate that embeds kaspa-pow, the node, the pool daemon (igneum-pool) and the app's CPU re-check alike, each built against the pinned igneum-pow by path from the release worktree (017e7037 on 0.3.22 and 0.3.23), and the packs pin travels with the generator (the proto-cuda packs are the pin's export, never an older one); the read-back is the paired miner's "class v4 program id <16 hex>" line equal to the node's per epoch, and the daemon accepting its shares. 8. **glibc classes:** HiveOS 2.31 (`--ship hive`, smoke in ubuntu:20.04 on the box), seeds and generic 2.35 (`--ship seed`), fleet 24.04 boxes native 2.39. From 7d4802c4fae3651e2e428d469375ec4f792efbcb Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:52:40 +0000 Subject: [PATCH 10/15] tools/ci/README.md: the scrubbed master copy with the inline-rm rule appended (the landing had carried a stale copy over the scrub) Co-Authored-By: Claude Fable 5.1 --- tools/ci/README.md | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/tools/ci/README.md b/tools/ci/README.md index 059771c39..82495fbc4 100644 --- a/tools/ci/README.md +++ b/tools/ci/README.md @@ -2,7 +2,12 @@ | Check | What it fails | Since | |---|---|---| -| no text overlaps (`overlap-check.mjs`) | A served page, or a miner or wallet screen (behind `IGNEUM_OVERLAP_APPS=1`), where a visible run of text is covered by another element (a pill over a caption, a label over a value, a card over its neighbour, text under the header at rest), clipped by an overflow-hidden ancestor, or past the viewport; a page that scrolls sideways. Five widths, light and dark, the home hero at rest and at each step. A fixture with one deliberate overlap of each kind must be flagged first (`--self-test`). Needs a headless Chromium: CI installs Playwright; the Mac ships the pages to build-2 (`infra/build-server/overlap-browser.sh`). | 7 October 2026: the hero's step pill sat on the caption's second line ("Two thirds of the weight sign. The checkpoint locks.") at every desktop width, found by the project lead on the live site | +| no text overlaps (`overlap-check.mjs`) | A served page, or a miner or wallet screen (behind `IGNEUM_OVERLAP_APPS=1`), where a visible run of text is covered by another element (a pill over a caption, a label over a value, a card over its neighbour, text under the header at rest), clipped by an overflow-hidden ancestor, or past the viewport; a page that scrolls sideways. Five widths, light and dark, the home hero at rest and at each step. A fixture with one deliberate overlap of each kind must be flagged first (`--self-test`). Needs a headless Chromium: CI installs Playwright; the Mac ships the pages to build-2 (`infra/build-server/overlap-browser.sh`). | 7 October 2026: the hero's step pill sat on the caption's second line ("Two thirds of the weight sign. The checkpoint locks.") at every desktop width, found by the founder on the live site | + +| master takes only CI-passed commits (`ci-state.mjs`, `merge-to-master.sh`, the hook's `master_ci_ok`) | A push to master whose commit, or whose merge's branch parent, has no green `ci` run on that exact sha (the runs API through gh: red, queued, none or gh unreachable all refuse); a merge onto a master whose last compiled run is red, unless declared the fix (`--fixes-master`). The merge tool pushes an unrun branch for a run and waits for a queued one with the clock. A feature-branch push prints the branch's previous red first (`--branch-red`). | 7 October 2026: era-vdf's tip 0e2d6b1c merged with no ci run; master's igneum-pow suite red from 16:31 UK under five docs-only green merges | +| every workflow job carries timeout-minutes (`workflow-timeouts-check.sh`) | A job in .github/workflows without `timeout-minutes`, or a budget off its measured line (site 15, changes 10, pow 60, sims 45). | 7 October 2026: three hosted site jobs on master hung over two hours each in the tree gate; the six-hour default was the only stop | +| a box or network check gets one retry (`retry-once.sh`) | Nothing by itself: wraps the box-locks check, the scene parity check and the live public API check so a first failure is printed and retried once; the second is the verdict. The checks keep their own skip line on a runner without the resource. | 7 October 2026 | +| the red watcher fires on cancelled and timed-out runs too (`ci-red.yml`, `red-watch.mjs`) | The watcher's `if` missing any of failure, cancelled, timed_out, or the conclusion not handed to the record step (the self-test reads the workflow file); the line names the kind: CI red, CI cancelled, CI timed out. | 7 October 2026 | | kill by exact command or pid file (owed as a check) | 6 October 2026, 21:09Z: a Mac-side `pkill -f ` matched nothing (the log name was a redirect, not part of the command line), the roll-everything script lived on and wiped a box it had been told to hold. Rule: a job is stopped by its pid file (`tools/fleet/fleet-bg.sh start|stop `) or by a pattern anchored on its exact command line (`^python3 -u /root/fleet/in/box-prover.py`), never by a word that may or may not appear in it. The check that flags a `pkill -f`/`pgrep -f` whose literal is a path or a name that never starts a command line is owed to the CI lane | From b732ab0b766283598354a691e18f2d9c58209401 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:54:20 +0000 Subject: [PATCH 11/15] release 0.3.22: decimals 18 and the testnet re-cut's line, the pre-place state for the one-minute move, dn3-g1 re-peered, one miner per card restored, the pool-mode seed-change finding, PC 1's helper cleared Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index d59906804..0fe4d6b59 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -126,3 +126,5 @@ Three commits on 96161037: bd710a36 the sub-version 3 re-pin (byte 7, igneum-pow ## 12. The 2720d8d2 split and the one-minute form (21:1x to 21:4x BST) The 0.3.23 node's heights move Devnet 3's digest to ba75bf6f, and the one-box-at-a-time sweep to it split the network: dn3-g1 flipped at 21:18 BST and sat alone twenty minutes on its own fork (blocks 12,553 to 12,793, peers 0; the 83eb50cd nodes refuse ba75bf6f at the handshake; build-1's seed logged 27 digest-mismatch rejects by 21:37); the fleet stopped the pass at 21:31 after that one box (the Vast ssh proxies ssh1, 3, 4, 5, 6 refused the next five, so nothing else flipped) and reverted dn3-g1 to 34a2dbaa on its kept datadir, where it reorgs onto the network's heavier chain; no record or share of the network's sat on the solo blocks; the pool pair, the provers and the second nodes stayed on 83eb50cd. The shipper's part of the fault: the 0.3.23 Mac entry (both folders, 21:39:09 BST, DMG 4ada2807 on 7c7489ac) and the 0.3.23 HiveOS alias (0d716ed9, 21:37) were published ahead of the fleet's move, against rule 5's own word; both pulled back to 0.3.22 (live again at 21:41:08 and 21:44:48 BST; the Mac still runs 0.3.22; the 0.3.23 hive package held in scratch r0323/held-public). The move to 2720d8d2 is the 0.3.20 form (rule 5 extended): binaries pre-placed on every Devnet 3 node (the fleet's thirty, build-1's three, the pool pair's two), every node restarted inside one minute at a clock the fleet names ten minutes ahead once the proxies answer, the digest read back per node, the first lock on the new side as the line; the 0.3.23 Mac entry, the hive alias and the Windows entry publish at that minute or after. Pool split: the hour's 13,209 + 7,200 = 20,409 lands before the 0.3.24 publish minute, so the split goes into the 0.3.24 object at or after the v5 floor on IGNH/IGNW (pending main); the node lane holds it out of the v5 object commit. PC 1: the 5090 floor grid exited 0 at 21:41 BST but the Power Helper hung at its 1,000 MHz step and the 5090 mines at the 1,100 MHz lock until the hash lane's unelevated restore job runs; the grid gets the fix (stop at the first helper timeout, restart the task before the reset). + +**Decimals (the founder, 21:40 BST): 18.** The testnet GO object re-cut goes on the 0.3.24 line as its second item on top of the v5 object commit (the 0.3.23 pin 2720d8d2 is frozen in the move): base_unit_decimals 18 with the UTXO/EVM scale fixed and tested, the 16-byte subsidy layout, the emission rescaled, the final 5 October message, the cache-rung field at 0, class v5 at 0 only if its Devnet 3 crossing reads clean, chain id 4462, the override file refused; the seeds re-armed on the dry run, nothing mining until the founder's go. **Pre-place for the one-minute move (the fleet, 21:5x BST):** the 2720d8d2 pair as .new on all 34 Devnet 3 nodes (20 rented, 14 standing second nodes), 2 of 34 placed while the Vast ssh proxies ssh1 to ssh9 refuse or time out since about 21:30; build-1's three pre-placed at 21:46 BST with the restart armed on the minute; dn3-g1 re-peered on 34a2dbaa at 21:43:32 BST (13 peers, three reorg lines, the 25 minutes of solo blocks orphaned). One miner per card restored by the fleet's v2 pass (p2-4090-3 first at 21:51:55 BST: the Devnet 3 miner alone at 55.96 MH/s, supervisors 4 to 1). **A pool-mode miner finding (0.3.24 item, the pool lane):** --exit-on-seed-change does not fire in pool mode; pool-b hashed the epoch-2 pack for 56 minutes after the 20:53 BST seed change (47,700 shares wrong_hash) until a re-export and restart by hand; the settled pool hour holds pool-b's shares to 20:53:09 only. PC 1: the Power Helper hang cleared by the hash lane's restore job at 21:45:58 BST (the 5090 back at 2,865 MHz); the 0.3.23 host slot open from 21:51 BST. From e6c2e0e9aca1fc3c0436a21702deefefdab227ee Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:56:25 +0000 Subject: [PATCH 12/15] Counter ASIC 3.0 status: the second 5090 pass (the knee at 1,300 MHz, the best points, the helper sequence fault and fix), the PC 1 queue, the derivation fix landed, class v5 frozen at dba17a0a with the kit line, the fourth restart-step exception and pair 4 Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index 5d6397dcb..b436da52c 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -455,7 +455,17 @@ Reading so far: the class v4 premium at the unlocked clock is +145 W on this car | 1,400 | 134.98 | 316.3 | 0.427 | 134.68 | 228.0 | 0.591 | 1,387 | | unlocked, end | 136.75 | 473.0 | | 136.5x | 329.7 | | 2,843 / 2,850 | -Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). THE FOUNDER'S QUESTION ON THE CLASS V4 PREMIUM ("we need a solution, deep research, other methods, something must be doable even if it is revolutionary"; a research lane, counter-asic-4, on the literature and the alternatives by 23:30Z). THE ONE COMPUTATION FROM THIS LANE (chip-model-v3 section 5.10, on the mirror's master at aa829826): class v5 ON and the shadow at ZERO leaves the strongest chip, the f = 1 stored-dataset chip, at 5.1x (GDDR7, the 5090's own 16 devices without the GPU: 166 MH/s at 78 W with a farm-shared node) to 9.1x (HBM3 eight stacks: 666 MH/s at 175 W) per joule over the 5090's 0.417 MH/W, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost (class-v5 2a.2: one node serves a farm, the leaves ship at 16.5 KB/s to 10,000 members, the rebuild is the same 32 ms per window every GPU pays); only a chip forced to carry its own node (85 W, approximate) falls near 2x, and only the small ones (one HBM3 stack 1.8x, the GDDR7 board 2.5x; the eight-stack package 6.2x). So the shadow stays the only lever in the model that reaches the memory-system chip, and the premium is its price; what class v5 buys is the recompute and stale chips gone as categories and every miner holding and following the chain. THE RESEARCH LANE's READING of the three shapes (21:4x UK; its identity edge = (E_card + F) / (E_mem + k F), F the GPU's premium per hash, k the chip core's energy per op over the GPU's): at the 1,400 MHz lock (v3 1.69 uJ per hash, v4 2.34, the premium 0.65 uJ = 6.5 pJ per counted op) the premium-free edge is 3.6x on GDDR7 and 5.3x on one HBM3 stack, 2.1x at F = 0.65 and k = 1, the asymptote 1/k; the premium needed for 2x at k = 1 is 0.76 uJ (103 W) at the lock and 1.41 uJ (175 W) at the stock point; at k at or under 0.5 no premium reaches 2x; a premium of zero is impossible by any hash-side lever (a joule the chip must spend is a joule the GPU spends first). Shape 1 (the shadow at the floor clock): correct and measured by the 1,400 row; it halves the premium and leaves k as the whole question (the edge's derivative in F at F = 0 is minus 5.7x per uJ at k = 1 and minus 0.2x per uJ at k = 0.3; the family mix with the highest k, shuffles and multiplies, is the right content; the tensor block has k near 1 but 0.056 pJ per MAC, so it forces no joules without 26x the verifier). Shape 2 (per-read work a chip cannot amortise): no construction found (the chip's lane count is set by its own latency, lane state is SRAM at pJ per access, extra reads scale both rates; row, bank, refresh and burst shaping have no asymmetry); one candidate for a measurement, not a claim: a hot table kept L2-resident through cache-policy hints (dataset loads evict-first, hot loads evict-last), since a 64 MiB SRAM read on a chip is not cheaper than a GPU L2 hit (the 3.0 layer 5 measurement used no hints and lost 13 to 16 percent). Shape 3 (the refresh as the cost): dead by arithmetic (a 1 GiB rebuild about 0.1 J on a chip against 280 kJ of hashing an hour). What lowers the premium-free floor is the card's own E_card: the 5090's idle 74 to 91 W plus its memory system's 55 W bound the floor near 2.0x on GDDR7 at any operating point; the two measurements that read how much of the 100 W between is reachable are the knee below 1,400 MHz (the floor pass, on PC 1's queue) and an SM-sparse kernel (the hash on a fraction of the SMs with 4 to 8 chains per thread, the rest clock-gated), both ordered to the hash lane with the L2 cache-policy hot table as PC 1 queue tail items (worker launch shapes, --bench only, no consensus change). THE FIVE RATIOS (adv-accept, 21:35 BST, at the acceptance's 2^20 sample, closed form and live to 1e-4): 4346 (bbb38e847011c354) minimum site 2 at 0.9840 REFUSED by (c''') at 0.995; 5245 (beaad44840bb9e4e) site 8 at 0.9831 REFUSED; 106924 at 0.9821 and 107022 at 0.9877 REFUSED; Devnet 3's epoch-0 class v4 program (fce15bf61030be57) site 0 at 0.9992 (live 0.9993) MISSED; era-fixed-20 (11f9f955b21d56c9) site 11 at 0.9997 MISSED; era-drawn-2 (7ceb797d31eedb3e) site 13 at 0.9992 MISSED. THE DEFENDER'S RULING: the (c''') floor at 0.995 closes the HOT-SET HALF of the class (every program with X_f at or above f that any lane measured, 100767 included) and NOT the class; the residual is the shadow-block-written value-level concentration below a ratio floor's resolution (1.26x to 1.45x of the top 0.1 percent share, 0.03 to 0.1 percent of a hash's reads each, minimum sites 0.9992 to 0.9997, which no floor reaches without sitting inside the clean seeds' own spread), chip gain under 1.001x, its lever a value-level source test at live scale or a per-site hot-item test, routed to the next class as a named item; section 14 of the v5 design says so with the five numbers verbatim; the freeze proceeds on the suite's green, both the floor's number and its reach measured. adv-mixer-2 (49656c2e) confirms AP-F4-1's class from outsider inputs (model A an FPGA LUT-area gain for a per-day build, 1.0 on every GPU, verifier and chip with a general multiplier; P(A at or above 1.1x) = 2^-10.8 per day on the exact median 226; model C under 2^-20 at 1.1x), the two censuses' medians (226 against 231) to be reconciled side by side in the record; its redraw rule to the v5 lane as a second independent statement. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh run by path under the inline-rm rule (v5-kits 4f9d96d2); the kit zip stands. MAIN'S RULINGS (21:4x UK). (1) The class v5 freeze does not wait on the three milder concentrations the 0.995 floor misses (Devnet 3's first program among them): their chip gain is bounded under 1.002x by their size, so the freeze proceeds with the floor as it is, the record names them as the residual, and the fix question (a value-level source test at live scale) is a 0.3.25 item. (2) The weak-day census reconciled at median 226 (F4's NAF weight had counted the carry digit at position 32, which a 32-bit multiplier never pays; 0.333 digits per word, 5.3 adders per day, the whole of 231 against 226 and of 12 against 15 days a century, the same worst day in both; F4's record section 9 carries both medians side by side, mirrored in adv-mixer-2's report): 5.69e-4 of days (2^-10.8), 15 days a century over 1.1x on FPGA LUT area, worst 2050-04-28 at 1.113x, the DSP-bound readings at 0 for k at least 2, F4's PASS against class v4 unchanged; every public text saying "12 days a century" becomes "15 days a century" (done in this landing on the litepaper table, /claims through it, evidence row 17 and the public text file, the sentence now "at most 12 percent more multiplier area on an FPGA's per-day build on 15 days a century, nothing on the other days and nothing for any chip"); the site audit lane told. (3) THE HALVING, ruled to the fleet (21:4x UK), neither (a) nor (b) as written: on every standing box the OLD shared-devnet miner stops and the Devnet 3 miner keeps the card at full rate (one miner per card, the exception over); hub-1, pool-1 and the live seed stay on the old chain as its voters and miners until every poller has moved; the old-chain nodes on the boxes stay up as voters without mining; no 3070 fallback; the fleet reports per-box restore lines. THE IN-HOUSE PASS, MORE CLOSES: adv-cache-3 COMPLETE (bc2d01d5, 0.23 slot-hours; every row BOUND or PASS, every plant fired; the exhaustive w = 2 image census at all 64 depths equal to the random-function recursion to 2 x 10^-5; the flip-table ladder at 2^18 lines no single-bit bias from two double rounds up). adv-accept-3 (aa359962): a correctness FINDING for the rule's owners, no chain consequence: the deterministic last-resort program (after 256 failed attempts) FAILS the real rule in 223 of 2,500 seeds (209 by part (a), a cyclic stale load) yet is handed to the chain unchecked, unreachable at 4.6e-44 per epoch; ruled to class v5's generator as the commit after the freeze (the last resort passing the rule by construction, known-failed first on one of the 223 seeds; sub-version 3's stated as unreachable and unverified, class v5's as verified); also the 256-unit stand-in ratio is noise as a selector (the three lowest of 4,975 clean live at 1.0002x worst), the selector working only at the 2^20-unit read. Lanes complete or at a natural end: adv-cache, adv-accept-2, adv-cache-3; adv-mixer on its last solve. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh by path (v5-kits 4f9d96d2) under the inline-rm rule. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of 6 October, which was read at the app's tuned cap) and 88.3 W at the 1,400 MHz lock; v4's rate is +0.18 percent over v3 unlocked and +0.23 percent at the lock; the rate is memory-bound on the whole grid (136.8 to 135.0 MH/s from 2,850 to 1,400); the best MH per watt sits at the lowest lock on the grid, so the knee is below 1,400 MHz. the founder's thesis holds in part: 57 W of the 145 W premium comes back by the lock alone; 88 W stays as the shadow's ALU work at the floor clock. Throttle reasons: the SW power-cap governor (0x400) from unlocked to 2,163, the lock itself (0x4) from 2,100 down on v4 and 1,650 down on v3; 75 C at the top, 59 C at 1,400. The owed 5090 clock rows (2,781 / 2,472 / 2,163 / 1,854) are in this table. Per tier: a 5090 owner on class v4 who locks the core at 1,400 MHz pays 316 W instead of 476 W for 1.4 percent less rate, MH per watt up 48 percent, the v4 premium down from 145 to 88 W; the lever is NVIDIA's -lgc through the helper; AMD has the helper's ADLX tune line or nothing; the Mac has no lever. MAIN'S ORDERS ON IT (20:2x UK): (1) the second pass now, 1,400 MHz down to the driver's floor in 100 MHz steps on v4 and the v3 control, to find the knee and the premium at it, then the same grid on the 5080; (2) the knob into Ember Tune for 0.3.23 (after the power-cap search, a core-clock search downward from the cap's point until the rate falls more than 1 percent, taking the best MH/W, the fingerprint on every step, stored per card; the Mac stated as no lever) through the UI lane; (3) the bench table's 5090 row gains the locked point and the class v4 cost column reads the locked premium beside the unlocked one (done: the 1,400 MHz row 134.98 MH/s at 316.3 W, 0.427 MH/W, the Hive values 1,400 / 13,801 / 575 as the driver's default limit). DEVNET 3's FIRST LOCK (the fleet lane): 19:02:46Z, checkpoint 235 LOCKED on block 50266abe... (blue score 7,050), identical on dn3-g1 (signed 98.4 percent of active) and dn3-g2 (100.1 percent), at DAA about 7,298 (the 7,200 window filled at 19:01Z); wave 1 of the second nodes GREEN on all five at 19:17:00Z, wave 2 from 19:17Z; the 0.3.22 pin candidate 34a2dbaa passed its last gate at 19:07Z (the join-and-restart read on dn3-c1 with the N15 line) and the Devnet 3 sweep to it runs. The fleet's 19:03Z table: 43 rows, 44 GPUs, USD 257 a day, today about USD 460 of the ceiling at 19:20Z; dn3-twin (a broken CUDA host) and dn3-q02 (never answered) destroyed; p12-vast, w-target and w-poison repurposed to Devnet 3. THE SITE: a0e0c83a deployed 19:17:28Z with the Hive column. THE BOXES: build-1 read 601 / 552 / 471 and build-2 401 / 417 / 400 at 19:17Z, the sum being adversarial binaries started by hand over ssh at 64 to 89 threads each (the crypto lane's one-sweep lock not holding the sum); MAIN'S RULE for every lane under this one: no run starts on a box except through the build-server lane's `lease pool -- cmd` (landing within the quarter hour); hand-started runs killed by their kill files and re-queued through the lease; the release builds and the v5 suites outrank the sweeps. THE SHIPPER cuts class v5 as 0.3.24 the minute every v5 gate is green, at any hour; this lane's gate board is the only clock. THE (c''') CENSUS NUMBER (the v5 lane, 21:03 UK; 4,600 f8 seeds, box 2, 48 cores, 1,256 s; log docs/design/class-v5-harness/v5-census-4600-0.log): the 0.995 per-site floor rejects 112 of 4,600 class v4 sub-version 3 accepted programs (2.435 percent); the same 112 move to a later class v5 attempt; the attempts mean 2.174 to 2.248 (+3.4 percent; the tail unchanged, max 24 on both); 0 class v5 accepted programs under the floor. The spread of the v4-accepted programs' minimum site ratio at the 2^20 sample: min 0.9807, p0.1 0.9831, p1 0.9906, p5 0.9962, median 0.9999; under 0.98 none, under 0.99 43 (0.935 percent), under 0.995 112, under 0.998 517, under 0.999 868. Two corrections to the relayed premise: the clean spread is not "0.9960 minimum, p1 0.9990" on the chain's own draw (a smaller sample), and 0.99 would NOT refuse the exemplar (0.9919). So 0.995 is the lowest round floor that refuses seed 100767 with the model's spread under it, at one extra draw attempt on 2.4 percent of epochs (about 0.3 s of acceptance each, no consensus cost, no change to the hash or the kits); the band it rejects is where every measured program so far is a weak hot-set program (adv-accept's live-low20 row: the first read, seed 3664 at 1.31x, beyond the 1.2x gate; its four measured lowest-ratio seeds all beyond). The per-site hot-item test is the same statistic at the 2^20 sample and costs 30 s per candidate at 2^24 against the floor's 0 extra, so it is not a competitor. The defender's call: keep 0.995; the freeze commit carries the number into accept.rs and section 14 and goes the minute the full suite and the gate read green (f17849eb staged). THE PRE-PUBLIC SCRUB: master's text pass (9b8eb23a, 3b4b6c63) names the founder as "the founder" in every tracked text file and the CI check founder-strings refuses the name; this record follows it from here (three lines of the efficiency pass reintroduced the name through a merge and are fixed). THE V5 FAST-TIME LINES ON THE FIRST MATCHED PAIR 959b57c9 (igneumd 519ee6c4..., igneum-pow ab6f980b; harness v5-fasttime 6de9cf74): (1) the ladder climb plus the v5 crossing with the stale node: rung 1 by miner signal from epoch 6 at 19:56:38Z; class v5 by signal from epoch 8 (DAA 480) at rung 1 at 19:58:31Z on 3 of 3 honest nodes, byte 6 at 9,985 bps, before the floor; the prelude reads v4 at epochs 0 and 1 (b1680b57 verified); honest nodes 0 PoW rejections; the stale node 122 of 122 refused by its own node; the SUMMARY FAIL at 20:04:00Z on line 3 alone. (2) Digest-compat SUMMARY PASS 20:08:30Z (one digest c0d6998e with the v5 key absent on the new binary and on the pre-v5 control 2720d8d2; the set key moves it to 8d8f6208 with no peer and the refusal line; 180 s of mixed mining, 79 new and 101 old blocks, 0 rejected). (3) THE RESTART STEP FAILED: n2 stopped at DAA 457 and restarted on its own datadir at DAA 500 (19:59:10Z), the IBD catch-up engaged and did not complete: the relay flow treated the engine's "class v5 needs the execution state" on a relayed epoch-8 block as a rule refusal twelve seconds in, the flow error tore down the peer's flows and the catch-up's body sync with them, the block went into the N6 refused memory, the peer was banned 600 s at 20:00:09Z, and the deferred headers were never validated; n2 on its own fork to the end (600 against 660). FIXED by the node lane at a3b2049d on class-v5-node-wire (20:0xZ): the relay flow holds off a block in the v5 state-wait class at its three validation sites (no strike, no N6 memory, the flow stays up; the next inv or the catch-up's deferred chunk brings it back once the executor has the state), is_v5_state_wait shared with the IBD catch-up, kaspa-p2p-flows 38 of 38 at 20:08:23Z. (4) One program id per epoch equal to the CLI's v5 id: 11 of 11 (v4 at rungs 0 and 1, v5 at rung 1 with the window's stream), the Devnet 3 genesis pack's e5a4ac5978462156 reproduced; the known-failed shape FAIL as it must at 19:04:01Z. THE SECOND MATCHED PAIR 63524e28 landed 20:16:49Z on build-1 (/srv/artefacts/v5-pair-63524e28/, igneumd sha256 4f93697f5b97c30205b7e1c8e521924ba85f849a5d35f1470ff0a1b2a17cbbae; fork v5-object-0323 = 959b57c9 + the relay hold-off a3b2049d + the pool lane's tag rename 7455b8d5 + main's chain id by height, Devnet 3 chain id 4464 from the v5 floor; igneum-pow ab6f980b; suites consensus 126, core 154, pow 19, exec 46, p2p-flows 38, miner 25); the fast-time lane re-runs the crossing with the restart step on it; the v5 object commit waits on that PASS, the freeze commit and dn3-g1's DAA at the cut. THE IN-HOUSE PASS: adv-accept's tally at 2^24: of the 16 lowest stand-in-ratio seeds, 9 beyond the 1.2x gate and 3 hot sets (100767 at 2.05x; 4346 at X/f 1.78, 2.24x; 5245 at 1.29, 1.86x); of 17 random accepted programs 1 beyond and 0 hot sets; the selector's false-positive rate 7 of 16 on the gate, 13 of 16 on the hot-set test; each hot set about 1 MB of items at 0.3 percent of reads, 1.002x; whether the 0.995 floor refuses 4346 and 5245 is a measurement in hand (their minimum-site ratios at 2^20), not a given. adv-accept-2 at its natural end (92168536): header grinding BOUND by measurement and by tail (1e8 hashes per real program on the windowed random baseline down to the 3e-7 tail, net gain 3e-7x); the rotate-identity repeat class present in 21.0 percent of 300 drawn programs at under 0.1 percent of loads on the worst, absent from the two real programs, a rule question for the next class. Lanes at a natural end: adv-cache, adv-accept-2. THE SITE: b8f501e9 deployed 20:17:42Z with the compact table; its 1600 px capture read five columns clipped at the article column's edge (about 900 px against a 1,400 px table), so the table now fits the column (nine columns, the generator in the detail row, the text columns wrapping and the numbers not, no forced width). THE FOUNDER'S QUESTION ON THE CLASS V4 PREMIUM ("we need a solution, deep research, other methods, something must be doable even if it is revolutionary"; a research lane, counter-asic-4, on the literature and the alternatives by 23:30Z). THE ONE COMPUTATION FROM THIS LANE (chip-model-v3 section 5.10, on the mirror's master at aa829826): class v5 ON and the shadow at ZERO leaves the strongest chip, the f = 1 stored-dataset chip, at 5.1x (GDDR7, the 5090's own 16 devices without the GPU: 166 MH/s at 78 W with a farm-shared node) to 9.1x (HBM3 eight stacks: 666 MH/s at 175 W) per joule over the 5090's 0.417 MH/W, the class v3 figures of 5.6 less a rounding, because the node is a farm cost and not a chip cost (class-v5 2a.2: one node serves a farm, the leaves ship at 16.5 KB/s to 10,000 members, the rebuild is the same 32 ms per window every GPU pays); only a chip forced to carry its own node (85 W, approximate) falls near 2x, and only the small ones (one HBM3 stack 1.8x, the GDDR7 board 2.5x; the eight-stack package 6.2x). So the shadow stays the only lever in the model that reaches the memory-system chip, and the premium is its price; what class v5 buys is the recompute and stale chips gone as categories and every miner holding and following the chain. THE RESEARCH LANE's READING of the three shapes (21:4x UK; its identity edge = (E_card + F) / (E_mem + k F), F the GPU's premium per hash, k the chip core's energy per op over the GPU's): at the 1,400 MHz lock (v3 1.69 uJ per hash, v4 2.34, the premium 0.65 uJ = 6.5 pJ per counted op) the premium-free edge is 3.6x on GDDR7 and 5.3x on one HBM3 stack, 2.1x at F = 0.65 and k = 1, the asymptote 1/k; the premium needed for 2x at k = 1 is 0.76 uJ (103 W) at the lock and 1.41 uJ (175 W) at the stock point; at k at or under 0.5 no premium reaches 2x; a premium of zero is impossible by any hash-side lever (a joule the chip must spend is a joule the GPU spends first). Shape 1 (the shadow at the floor clock): correct and measured by the 1,400 row; it halves the premium and leaves k as the whole question (the edge's derivative in F at F = 0 is minus 5.7x per uJ at k = 1 and minus 0.2x per uJ at k = 0.3; the family mix with the highest k, shuffles and multiplies, is the right content; the tensor block has k near 1 but 0.056 pJ per MAC, so it forces no joules without 26x the verifier). Shape 2 (per-read work a chip cannot amortise): no construction found (the chip's lane count is set by its own latency, lane state is SRAM at pJ per access, extra reads scale both rates; row, bank, refresh and burst shaping have no asymmetry); one candidate for a measurement, not a claim: a hot table kept L2-resident through cache-policy hints (dataset loads evict-first, hot loads evict-last), since a 64 MiB SRAM read on a chip is not cheaper than a GPU L2 hit (the 3.0 layer 5 measurement used no hints and lost 13 to 16 percent). Shape 3 (the refresh as the cost): dead by arithmetic (a 1 GiB rebuild about 0.1 J on a chip against 280 kJ of hashing an hour). What lowers the premium-free floor is the card's own E_card: the 5090's idle 74 to 91 W plus its memory system's 55 W bound the floor near 2.0x on GDDR7 at any operating point; the two measurements that read how much of the 100 W between is reachable are the knee below 1,400 MHz (the floor pass, on PC 1's queue) and an SM-sparse kernel (the hash on a fraction of the SMs with 4 to 8 chains per thread, the rest clock-gated), both ordered to the hash lane with the L2 cache-policy hot table as PC 1 queue tail items (worker launch shapes, --bench only, no consensus change). THE FIVE RATIOS (adv-accept, 21:35 BST, at the acceptance's 2^20 sample, closed form and live to 1e-4): 4346 (bbb38e847011c354) minimum site 2 at 0.9840 REFUSED by (c''') at 0.995; 5245 (beaad44840bb9e4e) site 8 at 0.9831 REFUSED; 106924 at 0.9821 and 107022 at 0.9877 REFUSED; Devnet 3's epoch-0 class v4 program (fce15bf61030be57) site 0 at 0.9992 (live 0.9993) MISSED; era-fixed-20 (11f9f955b21d56c9) site 11 at 0.9997 MISSED; era-drawn-2 (7ceb797d31eedb3e) site 13 at 0.9992 MISSED. THE DEFENDER'S RULING: the (c''') floor at 0.995 closes the HOT-SET HALF of the class (every program with X_f at or above f that any lane measured, 100767 included) and NOT the class; the residual is the shadow-block-written value-level concentration below a ratio floor's resolution (1.26x to 1.45x of the top 0.1 percent share, 0.03 to 0.1 percent of a hash's reads each, minimum sites 0.9992 to 0.9997, which no floor reaches without sitting inside the clean seeds' own spread), chip gain under 1.001x, its lever a value-level source test at live scale or a per-site hot-item test, routed to the next class as a named item; section 14 of the v5 design says so with the five numbers verbatim; the freeze proceeds on the suite's green, both the floor's number and its reach measured. adv-mixer-2 (49656c2e) confirms AP-F4-1's class from outsider inputs (model A an FPGA LUT-area gain for a per-day build, 1.0 on every GPU, verifier and chip with a general multiplier; P(A at or above 1.1x) = 2^-10.8 per day on the exact median 226; model C under 2^-20 at 1.1x), the two censuses' medians (226 against 231) to be reconciled side by side in the record; its redraw rule to the v5 lane as a second independent statement. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh run by path under the inline-rm rule (v5-kits 4f9d96d2); the kit zip stands. MAIN'S RULINGS (21:4x UK). (1) The class v5 freeze does not wait on the three milder concentrations the 0.995 floor misses (Devnet 3's first program among them): their chip gain is bounded under 1.002x by their size, so the freeze proceeds with the floor as it is, the record names them as the residual, and the fix question (a value-level source test at live scale) is a 0.3.25 item. (2) The weak-day census reconciled at median 226 (F4's NAF weight had counted the carry digit at position 32, which a 32-bit multiplier never pays; 0.333 digits per word, 5.3 adders per day, the whole of 231 against 226 and of 12 against 15 days a century, the same worst day in both; F4's record section 9 carries both medians side by side, mirrored in adv-mixer-2's report): 5.69e-4 of days (2^-10.8), 15 days a century over 1.1x on FPGA LUT area, worst 2050-04-28 at 1.113x, the DSP-bound readings at 0 for k at least 2, F4's PASS against class v4 unchanged; every public text saying "12 days a century" becomes "15 days a century" (done in this landing on the litepaper table, /claims through it, evidence row 17 and the public text file, the sentence now "at most 12 percent more multiplier area on an FPGA's per-day build on 15 days a century, nothing on the other days and nothing for any chip"); the site audit lane told. (3) THE HALVING, ruled to the fleet (21:4x UK), neither (a) nor (b) as written: on every standing box the OLD shared-devnet miner stops and the Devnet 3 miner keeps the card at full rate (one miner per card, the exception over); hub-1, pool-1 and the live seed stay on the old chain as its voters and miners until every poller has moved; the old-chain nodes on the boxes stay up as voters without mining; no 3070 fallback; the fleet reports per-box restore lines. THE IN-HOUSE PASS, MORE CLOSES: adv-cache-3 COMPLETE (bc2d01d5, 0.23 slot-hours; every row BOUND or PASS, every plant fired; the exhaustive w = 2 image census at all 64 depths equal to the random-function recursion to 2 x 10^-5; the flip-table ladder at 2^18 lines no single-bit bias from two double rounds up). adv-accept-3 (aa359962): a correctness FINDING for the rule's owners, no chain consequence: the deterministic last-resort program (after 256 failed attempts) FAILS the real rule in 223 of 2,500 seeds (209 by part (a), a cyclic stale load) yet is handed to the chain unchecked, unreachable at 4.6e-44 per epoch; ruled to class v5's generator as the commit after the freeze (the last resort passing the rule by construction, known-failed first on one of the 223 seeds; sub-version 3's stated as unreachable and unverified, class v5's as verified); also the 256-unit stand-in ratio is noise as a selector (the three lowest of 4,975 clean live at 1.0002x worst), the selector working only at the 2^20-unit read. Lanes complete or at a natural end: adv-cache, adv-accept-2, adv-cache-3; adv-mixer on its last solve. The kits lane's box-side body moved to tools/class-v5/kits-on-box.sh by path (v5-kits 4f9d96d2) under the inline-rm rule. THE SECOND 5090 PASS, THE KNEE (run-ca3-pc1-v4-eff-5090-floor-20261007, exit 0 at 20:41:14Z; the 5090 alone, 60 s steps, memory 13,801 MHz, every fingerprint matched): + +| Core lock MHz | v4 MH/s | v4 W | v4 MH/W | v3 MH/s | v3 W | v3 MH/W | sm read | +|---|---|---|---|---|---|---|---| +| unlocked | 136.84 | 473.7 | 0.289 | 136.50 | 329.9 | 0.414 | 2,842 / 2,850 | +| 1,400 | 135.02 | 320.0 | 0.422 | 134.85 | 229.0 | 0.589 | 1,387 | +| 1,300 | 134.76 | 312.5 | 0.431 | 134.62 | 223.3 | 0.603 | 1,290 | +| 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | +| 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | + +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From 9209d8a15425c4a851aa39c5e8434c860c76d2c7 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 20:59:56 +0000 Subject: [PATCH 13/15] Public chip texts: the X9 wording retired (its claimed ratio is against a CPU core); the floor and the premium as measured numbers at the 5090's knee (2.1x at k = 1, 3.4x at a core three times better, the premium 81.8 W, Ember Tune named); the ledger pins X35 and X36 moved; Counter ASIC 3.0 status: the research file's two orders Co-Authored-By: Claude Fable 5.1 --- docs/evidence.md | 2 +- docs/plans/counter-asic-3-public-text-2026-10-07.md | 10 +++++----- docs/plans/counter-asic-3-status.md | 2 +- site/claims.html | 2 +- site/evidence.html | 2 +- site/index.html | 2 +- site/litepaper.html | 10 +++++----- site/miner.html | 2 +- tools/ci/ledger-text-check.mjs | 6 +++--- 9 files changed, 19 insertions(+), 19 deletions(-) diff --git a/docs/evidence.md b/docs/evidence.md index 9cb19a950..2d5db9290 100644 --- a/docs/evidence.md +++ b/docs/evidence.md @@ -41,7 +41,7 @@ Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml` | 14 | Ethereum bytecode runs unchanged, with the documented differences of spec 7.1 | Homepage Build card; litepaper Building | tested by the team | as row 13; fixes `F-exec-A`, `F-exec-B` (spec 7.5) | `tools/evm-smoke/smoke.mjs`: deploy via viem, `increment`, `hashLoop`, `eth_estimateGas`, `eth_getLogs`; `tools/exec-attacks` scenarios 1 and 3; bench-log "execution layer attack fixes" | Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463; the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The `Prover` precompile, proof records and the shard planner are not in the node | none yet | | 15 | Every block is proven, with the proof landing within about a minute at launch | Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate | implemented | repo `d7e1f89` (GPU proof), `e01a3cc`, `292e800`, `eedd136` (`proving/igneum-prove`: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6 | `proving/windows-wsl2` (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; `igneum-prove-host --mode block` on `proving/fixtures/`; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards" | First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture `block-78-increment` (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in `docs/benchmarks/proving-e2e.md`. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on PC 2 in 34 s, verified on the Mac in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind `proving_v1_activation_daa` (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is one | none yet | | 16 | A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves) | Litepaper Proving ("The proving budget"); roadmap gate 2 | designed | spec 5.1 (Target), 7.6 (`S_p` provisional, 7,500,000 pgas = `B_p` / 4) | `PROVE-SHARD.bat` on the RTX 5090 (pending); the end-to-end standard in `docs/benchmarks/proving-e2e.md`; bench-log "proving: devnet v4 shards" | Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional `S_p` is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB card | none yet | -| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026). | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | +| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (a core as good as a GPU lane, k = 1) to 3.4x (a core three times better, k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 figure claimed against a CPU core and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.4x, 2.8x at launch; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026). | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | | 18 | The chip resistance measurements: the program is latency-bound (random reads), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache | Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page | tested by the team | readwidth e752fc7 (`docs/plans/read-width.md`), ca2-era 78c0ee4, ca2-cache 2de19e5 (`docs/plans/hot-table.md`) | The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per load | Latency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026 | none yet | | 19 | The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors | Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page | tested by the team | ca2-mixer 1ab8b21 (`tests/mixer.rs`, `tests/scratch.rs`), ca2-era 78c0ee4, ca2-soundness a465881 (`docs/analysis/scratch-soundness.md`), `igneum-pow/tests/packs.rs` | The crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per card | Class v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (PC 1 job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing) | none yet | | 20 | No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%) | Homepage stats and Economics tiles; litepaper Supply, Economics | implemented | repo `6ac80a3`; fork "igneum-node devnet v0"; `consensus/core/src/igneum.rs`, `coinbase.rs` | `cargo test -p kaspa-consensus-core igneum` (8 pass: subsidy table, ramp, split, cap) and `cargo test -p kaspa-consensus coinbase` (8 pass); `igneum-miner inspect 40`; bench-log "igneum-node devnet v0" | Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the `igneum-proving-pool-v0` output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happened | none yet | diff --git a/docs/plans/counter-asic-3-public-text-2026-10-07.md b/docs/plans/counter-asic-3-public-text-2026-10-07.md index 395f618f1..ee0afaded 100644 --- a/docs/plans/counter-asic-3-public-text-2026-10-07.md +++ b/docs/plans/counter-asic-3-public-text-2026-10-07.md @@ -1,10 +1,10 @@ -# The chip claim, public text (7 October 2026; REWRITTEN LAUNCH-FIRST 18:3x UK on the founder's "I thought we were making it 2.1 from launch?": the testnet and mainnet objects set program_class_v4_activation_daa to 0, so class v4 is live from genesis and the launch number is 2.1x to 3.9x on day one; the 5x to 9x is the class v3 baseline the work started from, stated only as that; the devnet's own activation height is a devnet fact only. Served since 11:03 UK on master 9162c847 with main's two cuts: no mention of the disclosure prize until the publish word, and row 17 in evidence.md's eight-column shape) +# The chip claim, public text (7 October 2026; REWRITTEN LAUNCH-FIRST 18:3x UK on the founder's "I thought we were making it 2.1 from launch?": the testnet and mainnet objects set program_class_v4_activation_daa to 0, so class v4 is live from genesis and the launch number is 2.1x to 3.4x on day one (2.1x with a core as good as a GPU lane, 3.4x with one three times better; the X9 wording retired 7 October 2026, 22:0x UK, on main's order: its claimed ratio was against a CPU core); the 5x to 9x is the class v3 baseline the work started from, stated only as that; the devnet's own activation height is a devnet fact only. Served since 11:03 UK on master 9162c847 with main's two cuts: no mention of the disclosure prize until the publish word, and row 17 in evidence.md's eight-column shape) Three texts and one ledger row, written by the Counter ASIC lane, which owns the chip model. Every number carries its label: measured (a card or a chain we ran, with the date), modelled (arithmetic on cited parts), claimed (a vendor's figure, never measured by us), designed (a rule in a class, not yet measured). Sources: `docs/analysis/chip-model-v3.md` sections 5 and 6, `docs/analysis/latency-shadow-2026-10-06.md`, `docs/plans/counter-asic-3-status.md`, `docs/analysis/attack-pass/f8-uniform.md` and `f4-weakday.md` (branch attack-pass), `docs/design/class-v5-stored-state.md`, the datacentre and market-cap rows of 7 October (lanes 3 and the fleet), the cryptanalysis plan in `docs/plans/funding.md`. ## 1. The home page's chip line (replaces the hero sentence served since 6 October 16:21Z) -Built for graphics cards. At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block. Class v5 then makes the dataset the chain's own state, so a chip that stores it or recomputes it is wrong on every item. Without class v4 the same chip would reach 5x to 9x. The model and every measurement are public. +Built for graphics cards. At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block; a 5090 locked at its knee pays 82 W for that shadow work. Class v5 then makes the dataset the chain's own state, so a chip that stores it or recomputes it is wrong on every item. Without class v4 the same chip would reach 5x to 9x. The model and every measurement are public. ## 2. The litepaper's chip section (replaces the paragraph that begins "The chip model: 5x to 9x per joule") @@ -12,7 +12,7 @@ The chip model. We price the strongest chip we can design against an RTX 5090 an | The chip and the class | Edge over an RTX 5090 per joule | Label and date | |---|---|---| -| At launch: a memory-controller chip that stores the whole dataset, under class v4 (about 100,000 integer ops per hash in the latency shadow, so the chip carries a GPU-class datapath beside its memory) | 2.1x with a core as costly per op as the GPU's (k = 1); 3.9x with the core Bitmain claimed for its Antminer X9 (k about 0.33), a product withdrawn before any unit shipped | modelled on measured card watts, 6 October 2026; the X9 figure claimed, never measured | +| At launch: a memory-controller chip that stores the whole dataset, under class v4 (about 100,000 integer ops per hash in the latency shadow, so the chip carries a GPU-class datapath beside its memory) | 2.1x with a core as costly per op as a GPU lane (k = 1); 3.4x with a core three times better per op (k about 0.33); no core below about 1.8 pJ per op is in the model's range, and the withdrawn Antminer X9's claimed figure is a ratio against a CPU core, not a GPU lane, so it is not a chip core against us | modelled on the 5090's measured watts at its knee, 7 October 2026 (the shadow's premium 81.8 W at the best points: class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W; a user gets there through Ember Tune's core-clock knob, 0.3.24) | | The same chip at the ladder's second rung (about 200,000 ops per hash), reached by miner signal | about 2.8x | modelled, 7 October 2026 | | Any chip under class v5, where the dataset is the chain's own state | a stateless or stale chip is wrong on every item, so the stored-dataset chip and the recompute chip are removed as categories; the verifier pays 0.2 ms more per warp | designed, 7 October 2026 | | A chip caching the hottest 0.1 percent of items (about 1 MB of SRAM) | bounded at 1.067x at the ceiling, 1.005x on about half the hours and 1.048x on 5 percent | measured census of 1,024 programs, 7 October 2026; the source rule in the next class | @@ -24,10 +24,10 @@ What a miner sees from this. Class v4 costs a 5090 about 80 W more for 0.2 perce ## 3. The miner page's line -Your card against the strongest chip we can price: an RTX 5090 at 136 MH/s on 350 W (measured 6 October 2026); at launch the chip reaches 2.1x to 3.9x per joule under class v4 (modelled on measured watts), and under class v5 it is wrong on every item because the dataset is the chain's own state (designed). Without class v4 it would be 5x to 9x. The model and the measurements are public. +Your card against the strongest chip we can price: an RTX 5090 at 136 MH/s on 350 W (measured 6 October 2026); at launch the chip reaches 2.1x per joule under class v4 with a core as good as a GPU lane, 3.4x with one three times better (modelled on measured watts at the 5090's knee: the shadow costs that card 82 W at its best point, and Ember Tune lands the lock by itself), and under class v5 it is wrong on every item because the dataset is the chain's own state (designed). Without class v4 it would be 5x to 9x. The model and the measurements are public. ## 4. The ledger row (docs/evidence.md row 17, in the table's eight columns as served) | # | Claim | Where it is made | Status | Version or commit | Reproducible test | Result, date, machine | Independent verification | |---|---|---|---|---|---|---|---| -| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | +| 17 | The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (a core as good as a GPU lane, k = 1) to 3.4x (a core three times better, k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state) | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 figure claimed against a CPU core and never measured | `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/plans/counter-asic-3-status.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; `docs/design/class-v5-stored-state.md`; the H100 and market-cap rows of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.4x, 2.8x at launch; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, PC 2's RTX 5090, PC 1's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word | diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index b436da52c..c8424f73f 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -465,7 +465,7 @@ Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of | 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | | 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | -The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule diff --git a/site/claims.html b/site/claims.html index af4758d91..5c057d780 100644 --- a/site/claims.html +++ b/site/claims.html @@ -222,7 +222,7 @@

    Here are the limits, stated before anyone else states them.

    • A proof in seconds. Not at launch. Proving a full block today needs a cluster of 100 to 200 consumer GPUs, approximate, so Igneum launches with proofs within about a minute and tightens as hardware improves. Users still see their transaction land in one second.
      • -
    • A chip is impossible. No. A chip wired for one program is a bad bet, because the program moves before it ships. A programmable chip is not stopped by the moving target: everything it needs is public at genesis and every drawn parameter is firmware to it (an address permute, a rotator, an immediate table), so the defence against it is the latency-shadow work (class v4) and the price per joule, not the schedule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block: a memory-controller chip that stores the whole dataset and carries a GPU-class datapath beside its memory for the 100,000 ops per hash in the shadow, the range running from a chip core as costly per operation as the GPU’s (k = 1, modelled on measured card watts, 6 October 2026) to the core Bitmain claimed for its withdrawn Antminer X9 (k about 0.33, never measured); the ladder’s second rung takes that bracket to about 2.8x (modelled, 7 October 2026). Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item (designed, 7 October 2026). The baseline the work started from, never the launch state: without class v4 the same stored-dataset chip would reach 1.2x per chip and 5x to 9x per joule in our model (6 October 2026); the Ethash chips of this class reached 2.1x to 4.8x (Linzhi Phoenix 2020, Jasminer X4 2021, Antminer E9 2022). The strongest recompute chip we can price, holding the whole 256 MiB cache on-die, reaches under 1x per chip against an RTX 5090 (the published model, 5 October 2026: 0.92x per unit of silicon with a 3x fixed-function allowance, approximate). Sources: the chip model analysis (6 October 2026); the ASIC history’s Ethash rows; Counter ASIC 3.0 item 8 (the chip’s per-joule edge over the RTX 5090 falls from 5.6x to 2.1x on GDDR7 at k = 1 and to 3.9x at the X9’s claimed core, the 5090 at 0.2% less rate; gates G1 to G6 passed, 6 October 2026). No hash has stayed free of chips forever; Igneum does not claim to. Monero’s RandomX has held for about seven years; the one chip announced against it, Bitmain’s Antminer X9, was withdrawn in mid-May 2026 before any unit shipped, its claimed core (k about 0.33) never measured. That record says nothing about the price of a chip with the 256 MB cache on its die; that price is a cost model, not a measurement.
      • +
    • A chip is impossible. No. A chip wired for one program is a bad bet, because the program moves before it ships. A programmable chip is not stopped by the moving target: everything it needs is public at genesis and every drawn parameter is firmware to it (an address permute, a rotator, an immediate table), so the defence against it is the latency-shadow work (class v4) and the price per joule, not the schedule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane and 3.4x with one three times better, under class v4 from the first block: a memory-controller chip that stores the whole dataset and carries a GPU-class datapath beside its memory for the 100,000 ops per hash in the shadow, the range running from a chip core as costly per operation as a GPU lane (k = 1, modelled on the 5090’s measured watts at its knee, 7 October 2026) to a core three times better per operation (k about 0.33); the withdrawn Antminer X9’s claimed figure is a ratio against a CPU core, not a GPU lane, so it does not stand for a chip core against us; the ladder’s second rung takes that bracket to about 2.8x (modelled, 7 October 2026). Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item (designed, 7 October 2026). The baseline the work started from, never the launch state: without class v4 the same stored-dataset chip would reach 1.2x per chip and 5x to 9x per joule in our model (6 October 2026); the Ethash chips of this class reached 2.1x to 4.8x (Linzhi Phoenix 2020, Jasminer X4 2021, Antminer E9 2022). The strongest recompute chip we can price, holding the whole 256 MiB cache on-die, reaches under 1x per chip against an RTX 5090 (the published model, 5 October 2026: 0.92x per unit of silicon with a 3x fixed-function allowance, approximate). Sources: the chip model analysis (6 October 2026); the ASIC history’s Ethash rows; Counter ASIC 3.0 item 8 (the chip’s per-joule edge over the RTX 5090 falls from 5.6x to 2.1x on GDDR7 at k = 1 and to 3.4x at a core three times better, the 5090 at 0.2% less rate; gates G1 to G6 passed, 6 October 2026). No hash has stayed free of chips forever; Igneum does not claim to. Monero’s RandomX has held for about seven years; the one chip announced against it, Bitmain’s Antminer X9, was withdrawn in mid-May 2026 before any unit shipped, its claimed core (k about 0.33) never measured. That record says nothing about the price of a chip with the 256 MB cache on its die; that price is a cost model, not a measurement.
    • A guaranteed income floor. No. External proving is a small market today. Igneum's miners' electricity cost in it is close to power, but the price they must charge is the subsidy they forgo, which falls as one over network hash: an edge at scale and nothing more.
    • A memory-hard prototype on every vendor. Not yet. The 256 MB cache closed the shortcut on Apple silicon (computing items runs 4.8x slower than loading them, measured 3 October 2026). The same ratio on NVIDIA and on a discrete AMD card is Open.
    • Finality in the first month. No. No checkpoint locks until the 30-day window has 30 days of history. The first month of mainnet is proof of work with a 12-hour depth, and the text above says so wherever a day count appears.
      • diff --git a/site/evidence.html b/site/evidence.html index a7fb3b72b..7a270cd2e 100644 --- a/site/evidence.html +++ b/site/evidence.html @@ -254,7 +254,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v 14Ethereum bytecode runs unchanged, with the documented differences of spec 7.1
      Homepage Build card; litepaper Building
      tested by the teamas row 13; fixes F-exec-A, F-exec-B (spec 7.5)tools/evm-smoke/smoke.mjs: deploy via viem, increment, hashLoop, eth_estimateGas, eth_getLogs; tools/exec-attacks scenarios 1 and 3; bench-log "execution layer attack fixes"Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463; the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The Prover precompile, proof records and the shard planner are not in the nodenone yet 15Every block is proven, with the proof landing within about a minute at launch
      Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate
      implementedrepo d7e1f89 (GPU proof), e01a3cc, 292e800, eedd136 (proving/igneum-prove: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6proving/windows-wsl2 (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; igneum-prove-host --mode block on proving/fixtures/; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards"First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture block-78-increment (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in docs/benchmarks/proving-e2e.md. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on the RTX 5090 Windows rig in 34 s, verified on the Apple M5 Max in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind proving_v1_activation_daa (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is onenone yet 16A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves)
      Litepaper Proving ("The proving budget"); roadmap gate 2
      designedspec 5.1 (Target), 7.6 (S_p provisional, 7,500,000 pgas = B_p / 4)PROVE-SHARD.bat on the RTX 5090 (pending); the end-to-end standard in docs/benchmarks/proving-e2e.md; bench-log "proving: devnet v4 shards"Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional S_p is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB cardnone yet -17The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (k = 1) to 3.9x (k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state)
      the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
      tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 core claimed and never measureddocs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/plans/counter-asic-3-status.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; docs/design/class-v5-stored-state.md; the H100 and market-cap rows of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.9x, 2.8x at launch; 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, the RTX 5090 Windows rig's RTX 5090, the three-card Windows rig's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026).none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word +17The chip resistance claim: at launch the strongest chip in the public model reaches 2.1x (a core as good as a GPU lane, k = 1) to 3.4x (a core three times better, k about 0.33) per joule against an RTX 5090 under class v4, live from genesis on the testnet and the mainnet; the ladder's second rung brings it to about 2.8x; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; a stored-dataset chip pays for itself only at about USD 100 M of market cap in two years; without class v4 the same chip would reach 5x to 9x (the class v3 baseline, the devnet's starting state, never the launch state)
      the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
      tested by the team (every card, the verifier, the two attack-pass bounds, the H100), the chip itself modelled, class v5 and the ladder designed, the X9 figure claimed against a CPU core and never measureddocs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/plans/counter-asic-3-status.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; docs/design/class-v5-stored-state.md; the H100 and market-cap rows of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the chip model's arithmetic in its file; the card rows by the benchmark package; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; 2.1x, 3.4x, 2.8x at launch; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; USD 100 M; 5.1x to 9.2x the class v3 baseline; 6 and 7 October 2026, the M5 Max, the RTX 5090 Windows rig's RTX 5090, the three-card Windows rig's RX 9070 XT and RTX 4070, a rented H100 SXM, igneum-build-1 The k about 0.33 bound is the implied core of Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025), withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: a claimed, unmeasured figure carried as the pessimistic bound, not a calibration point (attack pass AP-F5-1, 7 October 2026).none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), and the one outside check is staged and waits on its escrow and the publish word 18The chip resistance measurements: the program is latency-bound (random reads), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache
      Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page
      tested by the teamreadwidth e752fc7 (docs/plans/read-width.md), ca2-era 78c0ee4, ca2-cache 2de19e5 (docs/plans/hot-table.md)The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per loadLatency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026none yet 19The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors
      Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page
      tested by the teamca2-mixer 1ab8b21 (tests/mixer.rs, tests/scratch.rs), ca2-era 78c0ee4, ca2-soundness a465881 (docs/analysis/scratch-soundness.md), igneum-pow/tests/packs.rsThe crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per cardClass v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (the three-card Windows rig (RTX 5090, RTX 4070, RX 9070 XT) job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing)none yet 20No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%)
      Homepage stats and Economics tiles; litepaper Supply, Economics
      implementedrepo 6ac80a3; fork "igneum-node devnet v0"; consensus/core/src/igneum.rs, coinbase.rscargo test -p kaspa-consensus-core igneum (8 pass: subsidy table, ramp, split, cap) and cargo test -p kaspa-consensus coinbase (8 pass); igneum-miner inspect 40; bench-log "igneum-node devnet v0"Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the igneum-proving-pool-v0 output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happenednone yet diff --git a/site/index.html b/site/index.html index f645bcf18..c49f9a5ed 100644 --- a/site/index.html +++ b/site/index.html @@ -236,7 +236,7 @@
      01
      The same card finds the block and proves it.

      Both jobs pay. When you stop, the card still games.

      02
      A fair start.

      Nobody holds a coin before block one. The protocol carries no fee. The one payment to the project is the Ember software’s optional 1% dev fee, like other GPU miners, off with one flag.

      -
      03
      Built for graphics cards.

      At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block. Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item. Without class v4 the same chip would reach 5x to 9x. The model and every measurement are public.

      +
      03
      Built for graphics cards.

      At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block; a 5090 locked at its knee pays 82 W for that shadow work. Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item. Without class v4 the same chip would reach 5x to 9x. The model and every measurement are public.

    diff --git a/site/litepaper.html b/site/litepaper.html index 8a9b13fcc..8090dc7c5 100644 --- a/site/litepaper.html +++ b/site/litepaper.html @@ -314,7 +314,7 @@ body.all .pager{display:none}

    Abstract

    -

    Igneum is a proof-of-work blockchain built for graphics cards, where NVIDIA cards also prove every block with zero-knowledge proofs and sell proving to other chains. At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block; class v5 makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item; without class v4 the same chip would reach 5x to 9x: the chip model, every number labelled measured, modelled, claimed or designed.

    +

    Igneum is a proof-of-work blockchain built for graphics cards, where NVIDIA cards also prove every block with zero-knowledge proofs and sell proving to other chains. At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block; class v5 makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item; without class v4 the same chip would reach 5x to 9x: the chip model, every number labelled measured, modelled, claimed or designed.

    It runs the Ethereum virtual machine, so anything built for Ethereum runs on Igneum unchanged. Transactions are included in about one second, proven within about a minute at launch, and locked by miners within about two. There is no premine, no pre-sale, no treasury taken from emission, no stake anywhere in consensus, and no dependence on any other chain. Mining stays open to anyone with a GPU because the mining program changes every hour, so a chip built for one program is useless for the next, and a chip for the whole program space is a GPU without the graphics parts. No scheduled human release is needed to keep it that way. Writing new code, including an emergency fix to the proof system, is the one thing that takes a person, and it activates only on miner signalling.

    1 / s
    blocks, rising to 10
    @@ -437,11 +437,11 @@ body.all .pager{display:none}

    Three ideas carry the chip resistance. The hash rewrites itself. A new program every hour, drawn from the chain. Its memory pattern changes with it. The rules change on a schedule fixed at launch. No release, no vote. These are automatic schedule changes: they defeat a chip wired for one datapath and they need no human fork. Against a chip that stores the dataset every drawn parameter is firmware, and what meets that chip is the latency-shadow work (class v4) and the price per joule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). It waits on memory, not maths. Every hash is a chain of random reads into a table too big for a chip to carry. The wait is the same physics for everyone. Miners hold the switch. Spare defences are written into the rules, switched off. A miner signal turns one on, at the class-change threshold: miners signal three things at three thresholds, 60 percent of blue blocks over two weeks for a parameter genesis leaves open, 90 percent for an upgrade (new code), and 95 percent with a floor height for a class change. No fork.

    -

    The work that waits can grow. Class v4 adds a block of latency-shadow arithmetic to every hash, about 100,000 integer operations that run while the memory reads are in flight, so a chip that stores the whole dataset still has to pay for a core. That size sits on a ladder fixed at genesis, six rungs from about 100,000 to about 1,000,000 operations, and it moves one rung at a time only when 90 percent of blue blocks in each of seven consecutive days ask for it; it can never move two rungs inside a week and never past a rung the reference verifier cannot check under 10 ms with its sibling thread busy (measured on the build server, 6 October 2026: the first three rungs pass at 8.8, 8.9 and 9.2 ms, the fourth misses by 0.08 ms on a loaded box and stays out until a quiet re-measurement, the two doublings are out at 12.4 and 15.0 ms). What it buys, on the measured cards: against a dataset-storing chip whose core costs what an RTX 5090's does per operation, the chip's per-joule edge falls from 2.1x at the first rung to 1.3x at the third; against a core as good as the one Bitmain claimed for its withdrawn Antminer X9 (about 3x per joule over a desktop CPU, never measured), from 3.9x to 2.8x. What it costs, per rung, is measured too: the Apple tier gives up 3 points of rate at the first step and 6 more at the second, the RTX 5090 nothing until the second; so the miners who pay for a step are the ones who take it (ledger M34).

    +

    The work that waits can grow. Class v4 adds a block of latency-shadow arithmetic to every hash, about 100,000 integer operations that run while the memory reads are in flight, so a chip that stores the whole dataset still has to pay for a core. That size sits on a ladder fixed at genesis, six rungs from about 100,000 to about 1,000,000 operations, and it moves one rung at a time only when 90 percent of blue blocks in each of seven consecutive days ask for it; it can never move two rungs inside a week and never past a rung the reference verifier cannot check under 10 ms with its sibling thread busy (measured on the build server, 6 October 2026: the first three rungs pass at 8.8, 8.9 and 9.2 ms, the fourth misses by 0.08 ms on a loaded box and stays out until a quiet re-measurement, the two doublings are out at 12.4 and 15.0 ms). What it buys, on the measured cards: against a dataset-storing chip whose core costs what an RTX 5090's does per operation, the chip's per-joule edge falls from 2.1x at the first rung to 1.3x at the third; against a core as good as the one Bitmain claimed for its withdrawn Antminer X9 (about 3x per joule over a desktop CPU, never measured), from 3.4x to 2.8x. What it costs, per rung, is measured too: the Apple tier gives up 3 points of rate at the first step and 6 more at the second, the RTX 5090 nothing until the second; so the miners who pay for a step are the ones who take it (ledger M34).

    The chip model

    We price the strongest chip we can design against an RTX 5090 and publish the arithmetic. Class v4 is live from the first block on the testnet and the mainnet (the ladder’s rung 0 at genesis), so the launch number is the class v4 row. The honest card: an RTX 5090 mines class v3 at 136 MH/s on 350 W in the bench and 290 W in the app (measured, 6 October 2026); an Apple M5 Max at 27 MH/s on 21 W (measured, 6 October 2026); an H100 SXM at 249 MH/s, 98 percent of its random-read ceiling like the 5090, 1.78x the 5090’s hash at 1.15x the tuned 5090’s hash per watt and a third of the hash per rented dollar (measured, 7 October 2026), so datacentre silicon does not change the chip question. The CPU verifier takes 2.33 ms per warp of 32 hashes on one M5 Max core under class v4 (measured, 6 October 2026), against a gate of 10 ms.

    - + @@ -450,7 +450,7 @@ body.all .pager{display:none}
    The chip and the classEdge over an RTX 5090 per jouleLabel and date
    At launch: a memory-controller chip that stores the whole dataset, under class v4 (about 100,000 integer ops per hash in the latency shadow, so the chip carries a GPU-class datapath beside its memory)2.1x with a core as costly per op as the GPU’s (k = 1); 3.9x with the core Bitmain claimed for its Antminer X9 (k about 0.33), a product withdrawn before any unit shippedmodelled on measured card watts, 6 October 2026; the X9 figure claimed, never measured
    At launch: a memory-controller chip that stores the whole dataset, under class v4 (about 100,000 integer ops per hash in the latency shadow, so the chip carries a GPU-class datapath beside its memory)2.1x with a core as costly per op as a GPU lane (k = 1); 3.4x with a core three times better per op (k about 0.33); no core below about 1.8 pJ per op is in the model’s range, and the withdrawn Antminer X9’s claimed figure is a ratio against a CPU core, not a GPU lane, so it is not a chip core against usmodelled on the 5090’s measured watts at its knee, 7 October 2026 (the shadow’s premium 81.8 W at the best points: class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W; a user gets there through Ember Tune’s core-clock knob, 0.3.24)
    The same chip at the ladder’s second rung (about 200,000 ops per hash), reached by miner signalabout 2.8xmodelled, 7 October 2026
    Any chip under class v5, where the dataset is the chain’s own statea stateless or stale chip is wrong on every item, so the stored-dataset chip and the recompute chip are removed as categories; the verifier pays 0.2 ms more per warpdesigned, 7 October 2026
    A chip caching the hottest 0.1 percent of items (about 1 MB of SRAM)bounded at 1.067x at the ceiling, 1.005x on about half the hours and 1.048x on 5 percentmeasured census of 1,024 programs, 7 October 2026; the source rule in the next class
    The baseline the work started from: the same chip under class v3, without the shadow (the Ethash class)5x to 9x (5.1x on GDDR7, 9.2x on eight HBM3 stacks; the Ethash chips of this class reached 2.1x to 4.8x)modelled, 6 October 2026; the precedent measured by others, 2020 to 2022; never the launch state

    What a miner sees from this. Class v4 costs a 5090 about 80 W more for 0.2 percent of rate, an M5 Max 16 W more for 1.5 percent, an RX 9070 XT and an RTX 4070 nothing (all measured, 6 October 2026). The ladder that sets how much work rides in the shadow starts at rung 0 at genesis and climbs by miner signal; its third rung is inadmissible today because a server core verifies it in 10.85 ms, over the gate (measured, 7 October 2026). On the devnet, which started on class v3, class v4 arrives by miner signal at a published height (a devnet fact, not a launch one). The next test of the model is an internal adversarial pass, not an independent review: three lanes that have never worked on the hash code attack the mixer, the chained cache and the acceptance rule with only what an outsider has (the public kit, the frozen object, the spec, the harnesses) and publish the break or the bound they reach. The one outside check is staged and waits on its escrow and the publish word.

    -

    No hash has stayed free of chips forever. Igneum does not claim to. It states the gain its own model finds, the response takes a week, and both are measured. The model is public: the numbers; the claim is tested by paid independent cryptanalysis and the public benchmark. Monero has run on RandomX since 2019 (approximate) with no chip shipped. Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked. A box with about a 2x per joule edge over the best CPUs, and about 3x over a desktop, was withdrawn rather than face a RandomX re-tune of 1.5x or more. That is the band Igneum’s class v4 model sits in (2.1x to 3.9x over an RTX 5090), and the defence that held was a maintained algorithm with a credible upgrade path, which is what the ladder is.

    +

    No hash has stayed free of chips forever. Igneum does not claim to. It states the gain its own model finds, the response takes a week, and both are measured. The model is public: the numbers; the claim is tested by paid independent cryptanalysis and the public benchmark. Monero has run on RandomX since 2019 (approximate) with no chip shipped. Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked. A box with about a 2x per joule edge over the best CPUs, and about 3x over a desktop, was withdrawn rather than face a RandomX re-tune of 1.5x or more. That is the band Igneum’s class v4 model sits in (2.1x to 3.4x over an RTX 5090), and the defence that held was a maintained algorithm with a credible upgrade path, which is what the ladder is.

    One thing takes a person, here and on every chain that exists: writing new code. A chain cannot safely write its own generator, and it cannot safely tell a chip from a wave of honest new cards by hashrate alone. If the design above ever failed, anyone could publish a new generator and miners would switch it on by signalling, as Monero's community can fork. Igneum is built to make that day unlikely, and does not depend on avoiding it.

    @@ -806,7 +806,7 @@ body.all .pager{display:none}

    Here are the limits, stated before anyone else states them.

    • A proof in seconds. Not at launch. Proving a full block today needs a cluster of 100 to 200 consumer GPUs, approximate, so Igneum launches with proofs within about a minute and tightens as hardware improves. Users still see their transaction land in one second.
      • -
    • A chip is impossible. No. A chip wired for one program is a bad bet, because the program moves before it ships. A programmable chip is not stopped by the moving target: everything it needs is public at genesis and every drawn parameter is firmware to it (an address permute, a rotator, an immediate table), so the defence against it is the latency-shadow work (class v4) and the price per joule, not the schedule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block: a memory-controller chip that stores the whole dataset and carries a GPU-class datapath beside its memory for the 100,000 ops per hash in the shadow, the range running from a chip core as costly per operation as the GPU’s (k = 1, modelled on measured card watts, 6 October 2026) to the core Bitmain claimed for its withdrawn Antminer X9 (k about 0.33, never measured); the ladder’s second rung takes that bracket to about 2.8x (modelled, 7 October 2026). Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item (designed, 7 October 2026). The baseline the work started from, never the launch state: without class v4 the same stored-dataset chip would reach 1.2x per chip and 5x to 9x per joule in our model (6 October 2026); the Ethash chips of this class reached 2.1x to 4.8x (Linzhi Phoenix 2020, Jasminer X4 2021, Antminer E9 2022). The strongest recompute chip we can price, holding the whole 256 MiB cache on-die, reaches under 1x per chip against an RTX 5090 (the published model, 5 October 2026: 0.92x per unit of silicon with a 3x fixed-function allowance, approximate). Sources: the chip model analysis (6 October 2026); the ASIC history’s Ethash rows; Counter ASIC 3.0 item 8 (the chip’s per-joule edge over the RTX 5090 falls from 5.6x to 2.1x on GDDR7 at k = 1 and to 3.9x at the X9’s claimed core, the 5090 at 0.2% less rate; gates G1 to G6 passed, 6 October 2026). No hash has stayed free of chips forever; Igneum does not claim to. Monero’s RandomX has held for about seven years; the one chip announced against it, Bitmain’s Antminer X9, was withdrawn in mid-May 2026 before any unit shipped, its claimed core (k about 0.33) never measured. That record says nothing about the price of a chip with the 256 MB cache on its die; that price is a cost model, not a measurement.
      • +
    • A chip is impossible. No. A chip wired for one program is a bad bet, because the program moves before it ships. A programmable chip is not stopped by the moving target: everything it needs is public at genesis and every drawn parameter is firmware to it (an address permute, a rotator, an immediate table), so the defence against it is the latency-shadow work (class v4) and the price per joule, not the schedule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane and 3.4x with one three times better, under class v4 from the first block: a memory-controller chip that stores the whole dataset and carries a GPU-class datapath beside its memory for the 100,000 ops per hash in the shadow, the range running from a chip core as costly per operation as a GPU lane (k = 1, modelled on the 5090’s measured watts at its knee, 7 October 2026) to a core three times better per operation (k about 0.33); the withdrawn Antminer X9’s claimed figure is a ratio against a CPU core, not a GPU lane, so it does not stand for a chip core against us; the ladder’s second rung takes that bracket to about 2.8x (modelled, 7 October 2026). Class v5 then makes the dataset the chain’s own state, so a chip that stores it or recomputes it is wrong on every item (designed, 7 October 2026). The baseline the work started from, never the launch state: without class v4 the same stored-dataset chip would reach 1.2x per chip and 5x to 9x per joule in our model (6 October 2026); the Ethash chips of this class reached 2.1x to 4.8x (Linzhi Phoenix 2020, Jasminer X4 2021, Antminer E9 2022). The strongest recompute chip we can price, holding the whole 256 MiB cache on-die, reaches under 1x per chip against an RTX 5090 (the published model, 5 October 2026: 0.92x per unit of silicon with a 3x fixed-function allowance, approximate). Sources: the chip model analysis (6 October 2026); the ASIC history’s Ethash rows; Counter ASIC 3.0 item 8 (the chip’s per-joule edge over the RTX 5090 falls from 5.6x to 2.1x on GDDR7 at k = 1 and to 3.4x at a core three times better, the 5090 at 0.2% less rate; gates G1 to G6 passed, 6 October 2026). No hash has stayed free of chips forever; Igneum does not claim to. Monero’s RandomX has held for about seven years; the one chip announced against it, Bitmain’s Antminer X9, was withdrawn in mid-May 2026 before any unit shipped, its claimed core (k about 0.33) never measured. That record says nothing about the price of a chip with the 256 MB cache on its die; that price is a cost model, not a measurement.
    • A guaranteed income floor. No. External proving is a small market today. Igneum's miners' electricity cost in it is close to power, but the price they must charge is the subsidy they forgo, which falls as one over network hash: an edge at scale and nothing more.
    • A memory-hard prototype on every vendor. Not yet. The 256 MB cache closed the shortcut on Apple silicon (computing items runs 4.8x slower than loading them, measured 3 October 2026). The same ratio on NVIDIA and on a discrete AMD card is Open.
    • Finality in the first month. No. No checkpoint locks until the 30-day window has 30 days of history. The first month of mainnet is proof of work with a 12-hour depth, and the text above says so wherever a day count appears.
      • diff --git a/site/miner.html b/site/miner.html index 007731637..622d530e0 100644 --- a/site/miner.html +++ b/site/miner.html @@ -306,7 +306,7 @@ pre b{color:var(--molten-text);font-weight:500}

      Graphics cards only. A new mining program every hour, so no chip is built for it. 80% of every block to the card that finds it, 20% to the cards that prove it. No premine, no stake, no fee to any team. Every number above has a row in the bench table.

      -

      Your card against the strongest chip we can price: an RTX 5090 at 136 MH/s on 350 W (measured 6 October 2026); at launch the chip reaches 2.1x to 3.9x per joule under class v4 (modelled on measured watts), and under class v5 it is wrong on every item because the dataset is the chain’s own state (designed). Without class v4 it would be 5x to 9x. The model and the measurements are public.

      +

      Your card against the strongest chip we can price: an RTX 5090 at 136 MH/s on 350 W (measured 6 October 2026); at launch the chip reaches 2.1x per joule under class v4 with a core as good as a GPU lane, 3.4x with one three times better (modelled on measured watts at the 5090's knee: the shadow costs that card 82 W at its best point, and Ember Tune lands the lock by itself), and under class v5 it is wrong on every item because the dataset is the chain’s own state (designed). Without class v4 it would be 5x to 9x. The model and the measurements are public.

      diff --git a/tools/ci/ledger-text-check.mjs b/tools/ci/ledger-text-check.mjs index 09a97fae7..261889fbb 100644 --- a/tools/ci/ledger-text-check.mjs +++ b/tools/ci/ledger-text-check.mjs @@ -15,7 +15,7 @@ const REQUIRED = { ['X7', 'hello@igneum.network'], ['X31', 'The public testnet is weeks away: three seed nodes and the public RPC are up, and it opens when the go checklist closes.'], // 7 Oct 2026, 18:3x: the launch-first chip line of docs/plans/counter-asic-3-public-text-2026-10-07.md section 1 - ['X35', 'At launch the strongest chip in our public model reaches 2.1x to 3.9x per joule against an RTX 5090, under class v4 from the first block'], + ['X35', 'At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block'], ], 'litepaper.html': [ ['X3', 'Live rows arrive with the public testnet.'], @@ -26,8 +26,8 @@ const REQUIRED = { ['X34', 'was withdrawn in mid-May 2026 before any unit shipped; RandomX 2.0 shipped on 25 March 2026'], ['X36', 'Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked.'], ['X35', 'so the launch number is the class v4 row'], - ['X35', '3.9x with the core Bitmain claimed for its Antminer X9 (k about 0.33), a product withdrawn before any unit shipped'], - ['X36', 'the X9 figure claimed, never measured'], + ['X35', '3.4x with a core three times better per op (k about 0.33); no core below about 1.8 pJ per op is in the model’s range'], + ['X36', 'the withdrawn Antminer X9’s claimed figure is a ratio against a CPU core, not a GPU lane, so it is not a chip core against us'], ['M34', 'The work that waits can grow.'], ['M2', 'computing items on the fly runs 4.8x slower than loading them'], ['M4', 'ProgPoW, as KAWPOW on Ravencoin since 2020'], From f2da72ecd7048c7ccdab1b81273943dbbb8dde70 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 21:12:18 +0000 Subject: [PATCH 14/15] Counter ASIC 3.0 status: the 0.3.24 pairing ruled (dba17a0a; 19c60cf9 is 0.3.25's first row with the redraw rule and the verified last resort), the eighth hot set, the 5080 stock rows and the watts question Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index c8424f73f..48a1c0d6a 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -465,7 +465,7 @@ Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of | 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | | 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | -The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From 64477ff09831f02a542fa39b15cde82c68685b19 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Wed, 7 Oct 2026 21:15:14 +0000 Subject: [PATCH 15/15] Site: the bench table's v4 cost and tuned columns are one clause each (the sentences live in the detail row), so the five-number row fits the article box (the 22:10 BST capture read the cost column wrapping to six lines) Co-Authored-By: Claude Fable 5.1 --- site/build.mjs | 10 ++++++---- site/miner-bench.json | 2 +- site/miners.html | 14 +++++++------- 3 files changed, 14 insertions(+), 12 deletions(-) diff --git a/site/build.mjs b/site/build.mjs index b438f0b13..61a58d40a 100644 --- a/site/build.mjs +++ b/site/build.mjs @@ -431,10 +431,12 @@ for (const [file, active] of PAGES) { // with its data row on a sort. const heads = [ ['Card', 'card', 'text'], ['MH/s', 'mh_s', 'num'], ['W', 'watts', 'num'], ['MH/W', 'mh_per_w', 'num'], - ['Class v4 cost', 'v4_cost_short', 'text'], ['Tuned', 'tuned_short', 'text'], ['Hive core / mem / PL', 'hive_short', 'text'], ['Date', 'date', 'text'], ['Who', 'by', 'text'], + ['v4 cost', 'v4_cost_short', 'text'], ['Tuned', 'tuned_short', 'text'], ['Hive core / mem / PL', 'hive_short', 'text'], ['Date', 'date', 'text'], ['Who', 'by', 'text'], ]; - const shortV4 = (r) => { const v = r.v4_cost || 'not measured'; const m = v.match(/^([^;(]+?)(?:\s*[;(]|$)/); return m ? m[1].trim() : v; }; - const shortTuned = (r) => { const v = r.tuned || 'stock, mining'; return v.split(' (')[0]; }; + // the short forms are one clause: the first watts or percent figure of the cost ("+16 W", "+145.3 W unlocked"), the + // tune state's first words; the full sentences live in the detail row + const shortV4 = (r) => { const v = r.v4_cost || 'not measured'; if (/^not measured/.test(v)) return 'not measured'; const m = v.match(/^([+-]?[\d.,]+\s*(?:W|percent)(?:\s+(?:unlocked|of rate))?)/); return m ? m[1].replace(' of rate', ' rate') : v.split(/[,;(]| for | at /)[0].trim(); }; + const shortTuned = (r) => { const v = r.tuned || 'stock, mining'; return v.split(/[:;(]/)[0].replace('full Ember Tune', 'Ember Tune').replace('stock, bench only', 'stock, bench').replace('no lever on Apple silicon', 'no lever').trim(); }; const shortHive = (r) => { const h = r.hive; return (h && h.core_mhz != null) ? fmt(h.core_mhz) + ' / ' + fmt(h.mem_mhz) + ' / ' + fmt(h.pl_w) + ' W' : 'stock'; }; const cells = (r) => [ [r.card, r.card], [fmt(r.mh_s), r.mh_s], [r.watts == null ? 'not read' : fmt(r.watts), r.watts ?? -1], [r.mh_per_w == null ? 'not measured' : fmt3(r.mh_per_w), r.mh_per_w ?? -1], @@ -473,7 +475,7 @@ for (const [file, active] of PAGES) { }); })(); `; - const sortStyle = ''; + const sortStyle = ''; const table = render(cur, 'bench-current'); const earlierTable = render(earlier, 'bench-earlier'); // Ember Tune's fleet priors (site/miner-priors.json, tools/tuning.mjs --priors --site): one row per card model, diff --git a/site/miner-bench.json b/site/miner-bench.json index 0e06eb34f..cf1f19082 100644 --- a/site/miner-bench.json +++ b/site/miner-bench.json @@ -180,7 +180,7 @@ "source": "bench log: 6 October 2026, 16:01Z, Ember run 6 (ember-tune-pc1-6)", "by": "measured by the team", "note": "against 127.9 MH/s at 311.0 W untuned (0.411 MH/W): 84 W saved for 0.15 percent of rate; the ladder's floor, not yet its optimum", - "v4_cost": "as the row above", + "v4_cost": "not measured at this tune point (the Ember run was on the class v3 program); the unlocked and locked rows above carry the measured premium", "tuned": "full Ember Tune: 1,854 MHz core lock at the 100 percent cap", "driver_os": "NVIDIA driver, Windows 11", "hive": { diff --git a/site/miners.html b/site/miners.html index 24f1d41df..09f5a2f9b 100644 --- a/site/miners.html +++ b/site/miners.html @@ -158,7 +158,7 @@ table{min-width:560px}
      Download -
      DiscordGitHubReddit
      +
      DiscordGitHubReddit
      @@ -222,17 +222,17 @@ table{min-width:560px}
      -
      +

      The table

      One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. Click a column header to sort; the table opens by MH per watt. Integrated GPUs are not listed. The earlier classes sit below, collapsed.

      Why the rate fell from the first bench to today. The genesis program did 104 dependent random 4-byte loads per hash over a 1 GiB dataset; the hourly program and class v3 do 128, with the mixer between them; class v4 adds about 100,000 integer operations per hash that ride in the memory wait. So the hash is bound by random-read bandwidth by design, and a card's MH/s is a relative number: the difficulty follows it, and the same card earns the same share of blocks at 136 MH/s on class v3 as it did at 228 MH/s on the genesis program. What a miner compares is hash per watt, and what the chain cares about is the chip edge, which the shadow work is there to cut.

      -
      Apple M5 Max (40 GPU cores, Metal)      		27211.29+16 W for 1.5 percent of rate at 102,100 ops per hash, measured 6 October 2026no lever on Apple siliconstock2026-10-06team
      Generator: v2 · Class v4 cost: +16 W for 1.5 percent of rate at 102,100 ops per hash, measured 6 October 2026 · Tuned: no lever on Apple silicon (no clock or power control exposed); stock · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-miner Metal worker, class v3 program (macOS, Metal) · Source: Counter ASIC 3.0 status: item 8, the Apple M5 Max rows (IOReport GPU and DRAM watts) · Note: GPU plus DRAM watts, not wall
      NVIDIA H200 SXM (141 GB)      		313432.90.723not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA H100 SXM (80 GB)      		248.7385.60.645not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-cuda bench (Linux), class v3 control (NVIDIA driver 580.126.09, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep, a rented card) · Note: 424.0 W maximum; 98 percent of its random-read ceiling like the 5090; 1.78x the 5090's hash at 1.15x the tuned 5090's hash per watt and a third of the hash per rented dollar
      NVIDIA RTX 5090 (32 GB)      		127.7226.80.563as the row abovefull Ember Tune: 1,854 MHz core lock at the 100 percent cap1,854 / 13,801 / 460 W2026-10-06team
      Generator: v2 · Class v4 cost: as the row above · Tuned: full Ember Tune: 1,854 MHz core lock at the 100 percent cap · Hive flight sheet: core lock 1,854 MHz, mem 13,801 MHz, PL 460 W (measured 6 October 2026 (Ember run 6: the clock lock 1,854 MHz, the memory clock as read, the limit 460 W of 575 as the cap did not bind)) · Miner: Igneum Miner 0.3.13 + Ember Tune kit 6 (mining, class v3 program) (NVIDIA driver, Windows 11) · Source: bench log: 6 October 2026, 16:01Z, Ember run 6 (ember-tune-pc1-6) · Note: against 127.9 MH/s at 311.0 W untuned (0.411 MH/W): 84 W saved for 0.15 percent of rate; the ladder's floor, not yet its optimum
      NVIDIA RTX 5070 Ti (16 GB)      		78.4145.60.539not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA A100 SXM (80 GB)      		138.4266.30.52not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA A100 PCIe (80 GB)      		155299.60.517not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5070 (12 GB)      		52102.80.506not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5080 (16 GB)      		71.2143.40.496not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 148.6 W maximum; the team's own 5080 on a dock reads 60 MH/s warming and gets its full Ember Tune on 7 October
      NVIDIA B200 (180 GB)      		416.4855.60.487not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX PRO 6000 Blackwell (96 GB)      		130.5288.70.452not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5090 (32 GB)      		135316.30.427+88.3 W at this lock for +0.23 percent of rate, measured 7 October 2026core lock 1,400 MHz1,400 / 13,801 / 575 W2026-10-07team
      Generator: v2 · Class v4 cost: +88.3 W at this lock for +0.23 percent of rate, measured 7 October 2026 · Tuned: core lock 1,400 MHz (the efficiency pass's grid floor), memory 13,801 MHz, the driver's power limit untouched · Hive flight sheet: core lock 1,400 MHz, mem 13,801 MHz, PL 575 W (measured 7 October 2026 (the class v4 efficiency pass: the 1,400 MHz lock, the memory clock as read, the limit as the driver's default since the lock alone set the draw; the knee below 1,400 is the second pass's)) · Miner: igneum-worker-cuda bench (installed worker 0.3.20), class v4 program v4-devnet-epoch0 (NVIDIA driver 617.14, Windows 11) · Source: Counter ASIC 3.0 status: the class v4 efficiency pass (the efficiency pass job of 7 October 2026, 18:40 to 19:16 UTC, on the team's Windows desk machine, the core locked through the installed app's Power Helper task, no prompt) · Note: the v3 control at the same lock 134.68 MH/s at 228.0 W (0.591 MH/W); recovered 159.2 W for 1.36 percent of rate against the unlocked class v4 point; the best MH per watt on the grid, so the knee is below 1,400 MHz
      NVIDIA RTX 5060 (8 GB)      		31.375.40.415not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5090 (32 GB)      		136.13500.389+145.3 W at the unlocked corestock, bench onlystock2026-10-06team
      Generator: v2 · Class v4 cost: +145.3 W at the unlocked core (475.5 against 330.2 W) for +0.18 percent of rate; +88.3 W at the 1,400 MHz lock (316.3 against 228.0 W) for +0.23 percent; measured 7 October 2026 (the class v4 efficiency pass on the team's Windows desk machine, 60 s steps, every fingerprint matched) · Tuned: stock, bench only (unlocked core) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-cuda bench, class v3 program (mx8-devnet-epoch0) (NVIDIA driver, Windows 11) · Source: bench log: 6 October 2026, Counter ASIC 3.0 item 8, the 5090 rows (the control row) · Note: the control, unlocked; the 6 October +80 W reading was at the app's tuned cap; the rate is memory-bound from 2,850 to 1,400 MHz (136.8 to 135.0 MH/s), the best MH per watt at the lowest lock on the grid, so the knee is below 1,400 MHz (the second pass runs to the driver's floor)
      NVIDIA RTX 4070 (12 GB)      		3179.50.389+30 Wfull Ember Tune: 1,860 MHz core lock, 160 W cap1,863 / 10,251 / 100 W2026-10-06team
      Generator: v2 · Class v4 cost: +30 W (79 to 109 W) for +0.4 percent of rate at 102,100 ops per hash, measured 6 October 2026 · Tuned: full Ember Tune: 1,860 MHz core lock, 160 W cap (Ember run 6: 1,863 MHz at the 50 percent cap, 75.6 W) · Hive flight sheet: core lock 1,863 MHz, mem 10,251 MHz, PL 100 W (measured 6 October 2026 (Ember run 6: 1,863 MHz at the 50 percent cap of 200 W, 75.6 W drawn; the item 8 rows at 1,860 MHz and 160 W)) · Miner: igneum-worker-cuda bench (installed worker), class v3 control (NVIDIA driver, Windows 11) · Source: Counter ASIC 3.0 status: item 8, the RTX 4070 rows (job run-ca3-pc1-4070-shadow-20261006) · Note: 79.3 to 79.8 W at the tune point; 28.78 MH/s at 75.6 W (0.381) in Ember run 6 mining
      NVIDIA RTX 5090 (32 GB), fleet      		100.63080.327not measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) · Note: a standing voter's status beside its prover, 18:00Z; 122 MH/s at 308 W earlier in the day (0.396 MH/W); the team's own 5090 rows above
      NVIDIA RTX 4070 Ti (12 GB)      		31.3107.30.291not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 4090 (24 GB)      		52.3183.10.285not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, the RTX 4090 hands row · Note: the hands row: the card mines and proves (17.4 GB peak while proving); the standing 4090 voters read 44.5 to 50.8 MH/s this hour beside their provers
      NVIDIA RTX 5060 Ti (16 GB)      		30.9114.80.2690.1 percent of rate, measured 7 October 2026stock, bench onlystock2026-10-07team
      Generator: v2 · Class v4 cost: 0.1 percent of rate, measured 7 October 2026 · Tuned: stock, bench only (180 W default cap, never tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.19 package, prebuilt NVRTC worker (bench mode, class v4 program) (NVIDIA driver, Windows 11) · Source: bench log: 7 October 2026, the first 16 GB card: an RTX 5060 Ti in a Thunderbolt enclosure (run b) · Note: class v4 program, 128 loads per hash, 1 GiB dataset, 10-minute window on the card alone at the stock 180 W limit: 114.8 W mean; PCIe 4.0 x4 through the enclosure
      NVIDIA RTX 4060 Ti (8 GB)      		20.177.50.259not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3060 Ti (8 GB)      		33.1129.50.256not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3090 Ti (24 GB)      		62249.50.248not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3060 (12 GB)      		26.9111.60.241not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.126.09, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 114.5 W maximum; the Devnet 3 boxes read 23.7 to 25.7 MH/s beside their nodes
      NVIDIA L40S (48 GB)      		56.4240.70.234not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3080 Ti (12 GB)      		59267.30.221not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 283.0 W maximum
      NVIDIA RTX 3070 Ti (8 GB)      		39178.30.219not measuredstock, bench onlystock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 595.71.05, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      AMD Radeon RX 9070 XT (16 GB)      		18.91990.095+2 percent of ratestock, bench onlystock2026-10-06team
      Generator: v2 · Class v4 cost: +2 percent of rate (19.29 against 18.92 MH/s) at 102,100 ops per hash, watts owed, measured 6 October 2026 · Tuned: stock, bench only (the AMD tune pass runs 7 October: set only if the ADLX tune line reads, else measure-only) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-opencl bench (installed worker), class v3 control (Adrenalin 26.9.2, Windows 11) · Source: Counter ASIC 3.0 status: item 8, the RX 9070 XT rows (job run-ca3-pc1-amd-g1-shadow-20261006); watts from the app's telemetry read of 5 October 2026 (the job's ADLX sample parsed 0 rows) · Note: the card sits at 87 to 95 percent of its dependent random-read ceiling (2.42 to 2.68 G loads/s), in a Thunderbolt enclosure; AMD OpenCL 3683.0
      NVIDIA RTX 3090 (24 GB)      		50not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voters (status lines, 7 October 2026) · Note: the best of four standing voters this hour (42.2 to 50.0 MH/s) beside their provers; watts not read
      NVIDIA RTX A5000 (24 GB)      		47.6not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) · Note: a standing voter's status line, 18:00Z; watts not read
      NVIDIA RTX 3080 (10 GB)      		43.7not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) and the sweep's hands row · Note: a standing voter's status line, 18:00Z; the hands row of the sweep reads 40.82 MH/s at 204.9 W; 10 GB, no prover
      NVIDIA RTX 3070 (8 GB)      		33.7not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's 0.3.21 wipe canary (status line, 7 October 2026) · Note: the 0.3.21 wipe canary's status line, 17:07Z, beside its node; watts not read
      Intel Arc B580 (12 GB)      		11not readnot measured0.1 percent of rate, measured 7 October 2026stock, bench onlystock2026-10-07team
      Generator: v2 · Class v4 cost: 0.1 percent of rate, measured 7 October 2026 · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-opencl bench, class v4 program (Intel driver, Windows 11) · Source: bench log: 7 October 2026, the Intel Arc B580 (eGPU equals slot) · Note: the same rate in the Thunderbolt enclosure and the slot; watts not read on this run
      +
      Apple M5 Max (40 GPU cores, Metal)      		27211.29+16 Wno leverstock2026-10-06team
      Generator: v2 · Class v4 cost: +16 W for 1.5 percent of rate at 102,100 ops per hash, measured 6 October 2026 · Tuned: no lever on Apple silicon (no clock or power control exposed); stock · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-miner Metal worker, class v3 program (macOS, Metal) · Source: Counter ASIC 3.0 status: item 8, the Apple M5 Max rows (IOReport GPU and DRAM watts) · Note: GPU plus DRAM watts, not wall
      NVIDIA H200 SXM (141 GB)      		313432.90.723not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA H100 SXM (80 GB)      		248.7385.60.645not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-cuda bench (Linux), class v3 control (NVIDIA driver 580.126.09, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep, a rented card) · Note: 424.0 W maximum; 98 percent of its random-read ceiling like the 5090; 1.78x the 5090's hash at 1.15x the tuned 5090's hash per watt and a third of the hash per rented dollar
      NVIDIA RTX 5090 (32 GB)      		127.7226.80.563not measuredEmber Tune1,854 / 13,801 / 460 W2026-10-06team
      Generator: v2 · Class v4 cost: not measured at this tune point (the Ember run was on the class v3 program); the unlocked and locked rows above carry the measured premium · Tuned: full Ember Tune: 1,854 MHz core lock at the 100 percent cap · Hive flight sheet: core lock 1,854 MHz, mem 13,801 MHz, PL 460 W (measured 6 October 2026 (Ember run 6: the clock lock 1,854 MHz, the memory clock as read, the limit 460 W of 575 as the cap did not bind)) · Miner: Igneum Miner 0.3.13 + Ember Tune kit 6 (mining, class v3 program) (NVIDIA driver, Windows 11) · Source: bench log: 6 October 2026, 16:01Z, Ember run 6 (ember-tune-pc1-6) · Note: against 127.9 MH/s at 311.0 W untuned (0.411 MH/W): 84 W saved for 0.15 percent of rate; the ladder's floor, not yet its optimum
      NVIDIA RTX 5070 Ti (16 GB)      		78.4145.60.539not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA A100 SXM (80 GB)      		138.4266.30.52not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA A100 PCIe (80 GB)      		155299.60.517not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5070 (12 GB)      		52102.80.506not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5080 (16 GB)      		71.2143.40.496not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 148.6 W maximum; the team's own 5080 on a dock reads 60 MH/s warming and gets its full Ember Tune on 7 October
      NVIDIA B200 (180 GB)      		416.4855.60.487not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX PRO 6000 Blackwell (96 GB)      		130.5288.70.452not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5090 (32 GB)      		135316.30.427+88.3 Wcore lock 1,400 MHz1,400 / 13,801 / 575 W2026-10-07team
      Generator: v2 · Class v4 cost: +88.3 W at this lock for +0.23 percent of rate, measured 7 October 2026 · Tuned: core lock 1,400 MHz (the efficiency pass's grid floor), memory 13,801 MHz, the driver's power limit untouched · Hive flight sheet: core lock 1,400 MHz, mem 13,801 MHz, PL 575 W (measured 7 October 2026 (the class v4 efficiency pass: the 1,400 MHz lock, the memory clock as read, the limit as the driver's default since the lock alone set the draw; the knee below 1,400 is the second pass's)) · Miner: igneum-worker-cuda bench (installed worker 0.3.20), class v4 program v4-devnet-epoch0 (NVIDIA driver 617.14, Windows 11) · Source: Counter ASIC 3.0 status: the class v4 efficiency pass (the efficiency pass job of 7 October 2026, 18:40 to 19:16 UTC, on the team's Windows desk machine, the core locked through the installed app's Power Helper task, no prompt) · Note: the v3 control at the same lock 134.68 MH/s at 228.0 W (0.591 MH/W); recovered 159.2 W for 1.36 percent of rate against the unlocked class v4 point; the best MH per watt on the grid, so the knee is below 1,400 MHz
      NVIDIA RTX 5060 (8 GB)      		31.375.40.415not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 5090 (32 GB)      		136.13500.389+145.3 Wstock, benchstock2026-10-06team
      Generator: v2 · Class v4 cost: +145.3 W at the unlocked core (475.5 against 330.2 W) for +0.18 percent of rate; +88.3 W at the 1,400 MHz lock (316.3 against 228.0 W) for +0.23 percent; measured 7 October 2026 (the class v4 efficiency pass on the team's Windows desk machine, 60 s steps, every fingerprint matched) · Tuned: stock, bench only (unlocked core) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-cuda bench, class v3 program (mx8-devnet-epoch0) (NVIDIA driver, Windows 11) · Source: bench log: 6 October 2026, Counter ASIC 3.0 item 8, the 5090 rows (the control row) · Note: the control, unlocked; the 6 October +80 W reading was at the app's tuned cap; the rate is memory-bound from 2,850 to 1,400 MHz (136.8 to 135.0 MH/s), the best MH per watt at the lowest lock on the grid, so the knee is below 1,400 MHz (the second pass runs to the driver's floor)
      NVIDIA RTX 4070 (12 GB)      		3179.50.389+30 WEmber Tune1,863 / 10,251 / 100 W2026-10-06team
      Generator: v2 · Class v4 cost: +30 W (79 to 109 W) for +0.4 percent of rate at 102,100 ops per hash, measured 6 October 2026 · Tuned: full Ember Tune: 1,860 MHz core lock, 160 W cap (Ember run 6: 1,863 MHz at the 50 percent cap, 75.6 W) · Hive flight sheet: core lock 1,863 MHz, mem 10,251 MHz, PL 100 W (measured 6 October 2026 (Ember run 6: 1,863 MHz at the 50 percent cap of 200 W, 75.6 W drawn; the item 8 rows at 1,860 MHz and 160 W)) · Miner: igneum-worker-cuda bench (installed worker), class v3 control (NVIDIA driver, Windows 11) · Source: Counter ASIC 3.0 status: item 8, the RTX 4070 rows (job run-ca3-pc1-4070-shadow-20261006) · Note: 79.3 to 79.8 W at the tune point; 28.78 MH/s at 75.6 W (0.381) in Ember run 6 mining
      NVIDIA RTX 5090 (32 GB), fleet      		100.63080.327not measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) · Note: a standing voter's status beside its prover, 18:00Z; 122 MH/s at 308 W earlier in the day (0.396 MH/W); the team's own 5090 rows above
      NVIDIA RTX 4070 Ti (12 GB)      		31.3107.30.291not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 4090 (24 GB)      		52.3183.10.285not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, the RTX 4090 hands row · Note: the hands row: the card mines and proves (17.4 GB peak while proving); the standing 4090 voters read 44.5 to 50.8 MH/s this hour beside their provers
      NVIDIA RTX 5060 Ti (16 GB)      		30.9114.80.2690.1 percent ratestock, benchstock2026-10-07team
      Generator: v2 · Class v4 cost: 0.1 percent of rate, measured 7 October 2026 · Tuned: stock, bench only (180 W default cap, never tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.19 package, prebuilt NVRTC worker (bench mode, class v4 program) (NVIDIA driver, Windows 11) · Source: bench log: 7 October 2026, the first 16 GB card: an RTX 5060 Ti in a Thunderbolt enclosure (run b) · Note: class v4 program, 128 loads per hash, 1 GiB dataset, 10-minute window on the card alone at the stock 180 W limit: 114.8 W mean; PCIe 4.0 x4 through the enclosure
      NVIDIA RTX 4060 Ti (8 GB)      		20.177.50.259not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3060 Ti (8 GB)      		33.1129.50.256not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3090 Ti (24 GB)      		62249.50.248not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3060 (12 GB)      		26.9111.60.241not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.126.09, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 114.5 W maximum; the Devnet 3 boxes read 23.7 to 25.7 MH/s beside their nodes
      NVIDIA L40S (48 GB)      		56.4240.70.234not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      NVIDIA RTX 3080 Ti (12 GB)      		59267.30.221not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 580.65.06, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04) · Note: 283.0 W maximum
      NVIDIA RTX 3070 Ti (8 GB)      		39178.30.219not measuredstock, benchstock2026-10-07fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, bench only (rented, not tuned) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-worker-cuda, class v4 program (NVIDIA driver 595.71.05, Ubuntu 24.04) · Source: bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)
      AMD Radeon RX 9070 XT (16 GB)      		18.91990.095+2 percent ratestock, benchstock2026-10-06team
      Generator: v2 · Class v4 cost: +2 percent of rate (19.29 against 18.92 MH/s) at 102,100 ops per hash, watts owed, measured 6 October 2026 · Tuned: stock, bench only (the AMD tune pass runs 7 October: set only if the ADLX tune line reads, else measure-only) · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-opencl bench (installed worker), class v3 control (Adrenalin 26.9.2, Windows 11) · Source: Counter ASIC 3.0 status: item 8, the RX 9070 XT rows (job run-ca3-pc1-amd-g1-shadow-20261006); watts from the app's telemetry read of 5 October 2026 (the job's ADLX sample parsed 0 rows) · Note: the card sits at 87 to 95 percent of its dependent random-read ceiling (2.42 to 2.68 G loads/s), in a Thunderbolt enclosure; AMD OpenCL 3683.0
      NVIDIA RTX 3090 (24 GB)      		50not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voters (status lines, 7 October 2026) · Note: the best of four standing voters this hour (42.2 to 50.0 MH/s) beside their provers; watts not read
      NVIDIA RTX A5000 (24 GB)      		47.6not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) · Note: a standing voter's status line, 18:00Z; watts not read
      NVIDIA RTX 3080 (10 GB)      		43.7not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's standing voter (status line, 7 October 2026) and the sweep's hands row · Note: a standing voter's status line, 18:00Z; the hands row of the sweep reads 40.82 MH/s at 204.9 W; 10 GB, no prover
      NVIDIA RTX 3070 (8 GB)      		33.7not readnot measurednot measuredstock, miningstock2026-10-07reported, fleet
      Generator: v2 · Class v4 cost: not measured (class v4 program only) · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: hive package 0.3.20 igneum-miner, class v4 program (NVIDIA driver, Ubuntu 24.04) · Source: the fleet's 0.3.21 wipe canary (status line, 7 October 2026) · Note: the 0.3.21 wipe canary's status line, 17:07Z, beside its node; watts not read
      Intel Arc B580 (12 GB)      		11not readnot measured0.1 percent ratestock, benchstock2026-10-07team
      Generator: v2 · Class v4 cost: 0.1 percent of rate, measured 7 October 2026 · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-worker-opencl bench, class v4 program (Intel driver, Windows 11) · Source: bench log: 7 October 2026, the Intel Arc B580 (eGPU equals slot) · Note: the same rate in the Thunderbolt enclosure and the slot; watts not read on this run

      Rows on the current class: 32. Each row names the engineering log entry or the job it came from.

      The Hive flight sheet column. Where a card has a measured tune point, the column gives the core clock lock, the memory clock and the power limit to copy into a HiveOS flight sheet (core / mem / PL); the line under each row carries the label with the date, the class v4 cost in full, the miner and driver, the source and the note. Stock means no tune point has been measured yet. The Hive package mines at these settings through Hive's own overclock controls; the desktop app's Ember Tune lands on them by itself.

      Earlier classes (the genesis program, the hourly program, class v3 before the shadow): 6 rows, not comparable with the table above

      These rows are the bench numbers of 3 and 4 October 2026: the genesis program (104 loads per hash), the hourly program and the first class v3 miner. A higher MH/s here is a different hash, not a faster card.

      -
      Apple M5 Max (40 GPU cores, Metal)      		45.2not readnot measurednot measuredstock, bench onlystock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-metal bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090 first run (the Apple row of the same table) · Note: genesis program, 1 GiB dataset
      Apple M5 Max (40 GPU cores, Metal)      		26.7not readnot measurednot measuredstock, miningstock2026-10-04team
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-miner devnet v4, Metal worker with prepare · Source: bench log: 4 October 2026, first hourly program swap on the live devnet: compile-ahead, no pause, two cards · Note: live devnet v4, unbroken through the hour boundary
      Apple silicon laptop (model not reported)      		24.3not readnot measurednot measuredstock, miningstock2026-10-04reported, fleet
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.1 (DMG) · Source: bench log: 4 October 2026, first outside machine on the devnet: an Apple silicon laptop through the Igneum Miner app · Note: 21.0 MH/s average over 7 minutes, 24.3 MH/s at the moment of the report, 33 accepted blocks
      NVIDIA RTX 5090 (32 GB)      		229not readnot measurednot measuredstock, bench onlystock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-cuda bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090, memory-hard dataset (pack igneum-genesis-mh) · Note: genesis program, 104 loads per hash, 1 GiB dataset
      NVIDIA RTX 5090 (32 GB)      		185.3not readnot measurednot measuredstock, bench onlystock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-cuda bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090 first run, dataset sweep and second program · Note: hourly program, 128 loads per hash, 1 GiB dataset
      NVIDIA RTX 5090 (32 GB)      		124.2not readnot measurednot measuredstock, miningstock2026-10-04team
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.0 package, prebuilt NVRTC worker · Source: bench log: 4 October 2026, the gfx1036 worker fault and what the Apple M5 Max could and could not reproduce · Note: live devnet v4, 128 loads per hash, CPU re-check clean, 0 rejected
      +
      Apple M5 Max (40 GPU cores, Metal)      		45.2not readnot measurednot measuredstock, benchstock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-metal bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090 first run (the Apple row of the same table) · Note: genesis program, 1 GiB dataset
      Apple M5 Max (40 GPU cores, Metal)      		26.7not readnot measurednot measuredstock, miningstock2026-10-04team
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: igneum-miner devnet v4, Metal worker with prepare · Source: bench log: 4 October 2026, first hourly program swap on the live devnet: compile-ahead, no pause, two cards · Note: live devnet v4, unbroken through the hour boundary
      Apple silicon laptop (model not reported)      		24.3not readnot measurednot measuredstock, miningstock2026-10-04reported, fleet
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.1 (DMG) · Source: bench log: 4 October 2026, first outside machine on the devnet: an Apple silicon laptop through the Igneum Miner app · Note: 21.0 MH/s average over 7 minutes, 24.3 MH/s at the moment of the report, 33 accepted blocks
      NVIDIA RTX 5090 (32 GB)      		229not readnot measurednot measuredstock, benchstock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-cuda bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090, memory-hard dataset (pack igneum-genesis-mh) · Note: genesis program, 104 loads per hash, 1 GiB dataset
      NVIDIA RTX 5090 (32 GB)      		185.3not readnot measurednot measuredstock, benchstock2026-10-03team
      Generator: v1 · Class v4 cost: not measured · Tuned: stock, bench only · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: proto-cuda bench (prototype, not mining) · Source: bench log: 3 October 2026, RTX 5090 first run, dataset sweep and second program · Note: hourly program, 128 loads per hash, 1 GiB dataset
      NVIDIA RTX 5090 (32 GB)      		124.2not readnot measurednot measuredstock, miningstock2026-10-04team
      Generator: v2 · Class v4 cost: not measured · Tuned: stock, mining · Hive flight sheet: stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read) · Miner: Igneum Miner 0.3.0 package, prebuilt NVRTC worker · Source: bench log: 4 October 2026, the gfx1036 worker fault and what the Apple M5 Max could and could not reproduce · Note: live devnet v4, 128 loads per hash, CPU re-check clean, 0 rejected

      How a row gets here

      @@ -275,12 +275,12 @@ table{min-width:560px}

      Mined by GPUs. Proven by fire.

      WindowsmacOSLinuxHiveOS
      - Specification and test vectors + Specification and test vectors

      Run