igneum/docs/plans/release-0.3.21.md

# Release 0.3.21 (the cut after 0.3.20; staged 7 October 2026, 14:2x BST)

Tree: release-0.3.21 on origin, from release-0.3.20's closed tree (the 0.3.20 cut: pin c4459193, igneum-pow 8c728ca3 at byte 5, the floor-moved file). Rules: docs/plans/release-rules.md (4a every gate on every candidate from its build; 4b warm pods per gate class; 5 the sweep in waves).

## 1. What rides it (the project lead's list, 7 October 2026)

| Item | Owner | State |
|---|---|---|
| driver-check 1982dbc7 (the driver table measured: NVIDIA 617.42, AMD 26.9.2 with the amd.com referer, Intel 32.0.101.9034; the install thread with one elevated prompt; `--drivers` on the manifest) | the Intel lane | in: 39187097, afe4e2eb, 416d1f5d |
| miner-reliability, app half (miner-reliability-21 aa2bb07e, 18 commits: the readiness gate and the ladder, the load clock, the hold owner, the orphan sweep, the FAULT lines to the intake, the signed `cards` job kind, execrpc with `probe`; box 226 + 32 + 8, UI 49) | the reliability lane | merged into release-0.3.21 |
| miner-reliability, miner half (miner-reliability-20 f067f7c1: STATUS while waiting for a template, identities capped from the node's template time, the fetch back-off) | the node lane | staged on release-0.3.21-node at the sweep-end word |
| sub-version 2 (the hash lane's 07a809a7, byte 7, id a788661687db4bb3), pending the F8 census (about 15:00 BST; pass = all 64 seeds under 1.2x of the window model plus the suite and G1 lines); if it fails or slips past 20:00 BST the node ships byte 5 again with the re-pin dropped | the Counter lane, the node lane | held |
| the fork gate from horizon-node (6eb21fc9, six commits on finality.rs) | the node lane | dry-merged clean into c4459193 |
| the node's own rust-toolchain.toml | the node lane | in place on its worktree |
| the under-12 GB prove-instead switch (section 2) | the shipper | to write after the 0.3.20 publish |
| the per-architecture prover server (section 3) | the shipper, the fleet | to write after the 0.3.20 publish |
| pool-finish 434e8b9b (the pool crate: the TLS 1.3 member port with the authorize binding, the open pool, pool.md 10; fork pool-finish-node b0444f51 off 8097d600: the pool-mode miner, the IGNS/IGNP codec, pool_split_activation_daa at never on every network, so the digest does not move) and pool-mf-row 343dd83b (MF-11 in the register) | the pool lane | main's word 14:2x BST: rides 0.3.21 with the split switch at never; if it reds a 0.3.21 gate or costs more than one rebase round it moves to 0.3.22 |
| apps-harmony (the builder's redesign, Overview as the landing tab, the Prove switch fix), then gpu-logos on top of it, then ui-overlap-fixes, then scene-parity, in that order as they land, each with its own gate line | the UI lanes | to come |
| MF-11 (the app not returning after update-now; PC 2 silent since the 0.3.19 update-now at 10:34Z, its relay agent dead with it, so no job or task reaches it): the register row, the UI lane owning the cause, the relay agent as a service surviving the app as the restart path | the reliability lane, the UI lane, the relay lane | the row asked |
| the ids commit 55768f88 (`resolve_program_ids`: the override's fields, then the env or the host, then the embedded verifying keys; the exec suite 32 with the bare-node test known-failed first; built 13:35Z sha 279b1b690e854fc9): 0.3.21's FIRST node commit by main's word, with the ids gate (rule 4c) on every candidate | the node lane | on the mirror; its gates run under rule 4a |
| the 0.3.21 node order (main, 14:4x BST), each with its own gate line, one rebase round or 0.3.22: on c4459193, 55768f88; miner-reliability-20 f067f7c1 and the late-join commit 70e4601e; pool-finish-node b0444f51; horizon-node 6eb21fc9 (the deep fork-choice fix, fork_gate never set); peer-directory-node db28d331 (the peer directory prototype, peer_directory_activation_daa at never, the IGNP announce only under --announce); the sub-version-2 re-pin held for the census. The horizon lane rebases its two onto c4459193 at the sweep-end word | the node lane, the horizon lane | staged tonight |
| pool-finish-21 at 5e557171 on origin (release-0.3.21 5d153892 plus the ten pool commits, the tenth d5265fb3 the `--template-parallel` fix, default 2: the ten-member window's daemon queued ten template calls on one gRPC connection past its own timeout, 1,891 lines, no job for 36 minutes; site/miner.html carries the pool-fee sentence); igneum-pool 28 at 5e557171 on build-2, the app gate 226 + 32 + 8 at 54dd4c06 (the last commit touches pool/ and docs only); merged after the three UI branches with the app gate rerun at the merge | the pool lane | ready to merge |
| 0.3.20's version strings moved to 0.3.21 | the shipper | 0d7b9bd0 |
| master 819d536b merged (the 51-check gate; build-remote routes by load; the hands script's pgrep in the bracket form) | the shipper | c83ca904, 0b75cf52 |

## 2. The under-12 GB prove-instead switch (design, main's routing)

The fact (the fleet's 3080 hour, 7 October 2026): a 10 GB card completes the compressed step alone at the default threshold (8,642 to 8,729 MiB of 9,885) and never beside its miner's 1,547 MiB; no lower threshold fits (the server dies before the compressed step at 524288 and below). 0.3.20 ships the rule "proving needs a 12 GB card; mining continues" (provedefault::prove_refused_under_12gb). 0.3.21 adds the choice: a Settings switch `prove_instead` (off by default) that, on a machine whose only NVIDIA card is under 12 GB, stops that card's miner while the prover runs and restarts it when the prover is off; the tile sentence becomes "proving instead of mining on <card> (10 GB holds one, not both)"; the refusal line stays when the switch is off. Engine: the prover loop's refusal rung reads the switch; when on, it takes the miners hold for that card (the hold owner from miner-reliability) for the prover's lifetime. Test: the refusal lifts with the switch on and the card's miner reads held; a 12 GB card beside it never triggers the hold. UI: the switch under Settings > Proving with the sentence; the view test.

## 3. The per-architecture prover server (design, main's rule)

The fact (the fleet, 7 October 2026): sp1-gpu-server is built for one card's compute capability (sm_86 Ampere, sm_89 Ada, sm_120 Blackwell); a server for the wrong architecture fails every proof in 12 s with "CudaRustError: named symbol not found" and the miner never notices; the 610-series driver proves (the driver floor stays 570 or newer, no upper bound). The app's WSL2 setup (setup-wsl.sh) builds the server on the machine, so it matches by construction; the risk is a copied or stale server. 0.3.21: (1) the prover's setup records the compute capability the server was built for (nvidia-smi --query-gpu=compute_cap) beside the binary; (2) at every prover start the engine reads the card's compute_cap and the record, and refuses with "the prover was built for sm_NN; this card is sm_MM; Set up rebuilds it" when they differ (MF-10's capability check in prover.rs, owed by the reliability lane's register); (3) the fleet's kit ships one server per architecture under bin-<arch> picked by compute_cap at install (the fleet lane). Test: a record and a card that differ refuse with the sentence; equal ones pass; no record passes with a warning line.

## 4. Gate clock (rules 4a and 4b)

App: the gate on build-2 about 19:30 BST (`build-remote.sh --box 2 -- test --release` from app/igneum-app), the UI tests, the Windows cross on build-2, the reliability injector's eight steps on a one-shot pod (the fleet, about 45 minutes, after the sweep). Node: the branch commits at the sweep-end word (about 16:45 BST), suites on build-2 about 16:50 to 17:05, the first candidate binary on build-1 about 17:10 with the digest gate, the mixed-version gate, the kept start, the cases and the wipe all started from the build on the warm pods; the pin about 80 minutes after the last candidate builds. The Mac's own binaries and the DMG under the lock after the pin; the staging in a scratch copy; the publish on green with the clock time.

**gpu-logos-21 in (14:4x BST):** ba5d9cea, four commits on 8ed08dcf (1054616c the vendor marks, bca70dcd the captures, c98f2666 the Prove switch fix as its own commit: the DAG legend's bare .lg and the inspector's bare .track scoped to .legend .lg and #i-track, the switch's size class its own, the native input hidden the accessible way under a positioned label; ba5d9cea the shared marks module for the site and the re-taken shots). Lines: build-2 app gate 226 + 32 + 8 at c98f2666 (ba5d9cea touches nothing in the crate); UI 72 pass (view 41: 5 marks, 3 switch known-failed first, 1 module drift); pre-push 52 green. Captures docs/plans/gpu-logos-shots/ 01 to 12. Merged into release-0.3.21.

**earnings-tidy-21 in (14:4x BST):** a436c6f0 on 5d153892 (c7ec4573 the Earnings card tidied by the project lead's order "remove all the type a price stuff": the £ per IGN input, Use it, the remembered price, the price, balance-in-pounds, share, IGN per kWh and cost-of-hash lines gone; the card is the day rate with its reason, blocks and IGN this run, a row of three (weight, electricity, lifetime), the dev-fee switch; View.earningsLines(s, chain, pence, now, minor), FIELDS.price gone; a436c6f0 the captures docs/plans/earnings-tidy-shots/01 to 03). Lines: build-2 app gate 226 + 32 + 8; UI 75 (three new, known-failed first); pre-push 52. Merged into release-0.3.21.

**miner-reliability-21 at cbd6f3a4 in (14:5x BST):** four more commits merged: 4a28eb59 MF-10 (provedefault::server_mismatch and is_arch_failure, the test sm_86 on 8.9 and 12.0 refused, 8.6 and a fat binary pass; prover.rs server_arch_check reads the card's compute_cap and the server's sm_ words at the first probe, refuses the GPU path with the reason on the tile and a FAULT class=prover-arch line; "named symbol not found" logged as the same class) and MF-8's injector step (app-run.mjs --node-old: the previous release's node writes 150 blocks and stops, the new node opens the datadir and reads synced in one start); a751d391, 6a51c6c1, cbd6f3a4 the register (MF-11 the machine going silent with nothing reporting it, MF-12 the pool member stalled at an epoch boundary carried over from the pool lane, the rows moved from owed to code). Lines: build-2 app tests 227 + 32 + 8 on 4a28eb59; pre-push 52; UI 49 (the lane's run; the tree's UI tests re-run at the next merge). The 0.3.21 injector run: the payload staged (the app 786c3d37 from 8ed08dcf, node c4459193, miner 65d0837a, the old 0.3.17-line igneumd for --node-old), nine steps about 50 minutes on the fleet's pod after the sweep. This closes section 3's app half (the capability check); the per-architecture kit is the fleet's.

## 5. The node line's first candidate: 55768f88 (the node lane, 14:5x BST)

sha256 279b1b690e854fc9, the string read back, pairing 8c728ca3 at byte 5. The digest gate 13:35:41 to 13:37:19Z PASS (thirteen fields a89be8a7 on both binaries, sixteen fields db9a85f9 refused with no peer, the live file's digest eada4bda unmoved); the ten-minute mixed-version gate beside the 5899f603 pair 13:37:40 to 13:47:52Z PASS (digest b0afb2ee on all five, 223 new and 381 old blocks accepted, 0 rejected, plain header version 2, counts equal at 319, 486 and 604 through both joins and the restart step, no panic). The fleet's set on the same binary (the kept start with the ids gate, the cases on the warm set, the wipe) runs under rule 4a. Staging at the sweep-end word in main's order: f067f7c1 and 70e4601e, then b0444f51, then the horizon lane's rebased 6eb21fc9 and db28d331, then the re-pin on the Counter lane's word; suites on build-2 and the digest read after every merge; the next candidate's binary with every gate from its build.

**scene-parity-21 in (14:5x BST), ahead of ui-overlap-fixes-21 in the order because it was green on the exact tip (50ffa562) while ui-overlap-fixes still rebases:** 20c9153b (0d75bc1f the shared scene/ folder and live-dag.js 2.0.3, c1334faf the app side, 20c9153b the parity harness and its gate line). Lines: build-2 app gate 228 + 32 + 8; UI 72; pre-push 56 with four new chain-scene checks (sync byte-equal, the paint-on-push known-failed test, the feed contract, the parity render on build-2: home fold = /live = app Inspect at T+0, +2, +4 s). live-dag.js 2.0.2 → 2.0.3 (paint on every push whatever the visibility, the blank /live fix; the phone rule on the viewport width); proof-core.js unchanged 2.0.0; the site's copies moved on master f7743534 (igneum.network/live-dag.js reads 2.0.3); the source is scene/live-dag.js, both copies written by `node tools/scene/sync.mjs`, the gate refuses drift, so the 0.3.21 pack carries the 2.0.3 bytes. Users see: the light theme's ember at the brand package's #D0420D, the chain feed asking 300 s, Inspect 420 px tall (seven lanes), the phone layout no longer frozen at launch width, api/live keeping the key id in `miner`. Captures: build-2 /srv/builds/scene-parity/igneum-wt-scene-parity/_out and scratchpad/scene/parity-out.

## 6. The shape that ships: the 0.3.21 app tree over the field node c4459193 (main, 17:5x BST)

No node gate for 0.3.21: the app tree ships over the node already in the field (c4459193, the 0.3.20 pin, digest 4bbbe816 on the floor file). The 96161037 line (N15's numbering class, the bare-node proving ids) folds into 0.3.22 and later; its canary cells (kept-datadir ids BOTH PRESENT, wipe canary c22-1 synced in 42 minutes, 23 for 23, restart synced in 1 min 24 s) stand as readings, not as this release's gate.

**The exact tree:** release-0.3.21 at 44b63ac9 = scaling-21 b340b904 merged (c999a104), the windows.yml smoke fix 6c4686e7 (a direct call with the exit code read, in place of Start-Process -Wait -PassThru, which raced the version read-back on run 37653903394), and the node-source pin back to c4459193 (44b63ac9). App gate GREEN on build-2 on that tree at 18:05 BST: 257 + 32 + 8, rc 0. Payload inputs on the dl host at 18:04 BST: igneumd.exe 49502cc7, igneum-miner.exe b4871b5d (c4459193's Windows pair, the 0.3.20 release's bytes), igneum-worker-cuda.exe d7a413c7, igneum-worker-opencl-intel.exe af53f194, igneum-gpu-telemetry 0d68d06e, the Linux prover pair.

**The GitHub exception window, from 18:02 BST:** GitHub answers "Your account was suspended" (403) to every call from the igneum-labs login, API and git; the windows.yml dispatch at 18:04 BST was that 403 and no run started. Main's ruling: no lane pushes, fetches or polls GitHub, not even a retry; the box mirror /srv/igneum.git on build-1 and build-2 is origin for every lane; the box gate stamp replaces merge-to-master.sh's CI wait; the window and its start are recorded here and closed by a row when the login is restored. release-0.3.21 44b63ac9 reached both mirrors at 18:12 BST; master a4bca198 was fast-forwarded onto the mirrors (they had sat at 83977817) so every lane's origin carries a current master.

**Windows without windows.yml:** the 0.3.10 shape. igneum-app.exe cross-built on build-1 from 44b63ac9 (GNU target, 151803c7, 4,232,704 B; igneum-ota-sign.exe ddfd9444; igneum-prove-verify.exe dbda5323), the payload zip igneum-windows-app-0.3.21-44b63ac9.zip 586beedd and the installer kit installer-kit-44b63ac9.zip 6d0b5437 on the dl host; the window host (MSVC, WebView2) and the installer (Inno Setup, rcedit) only build on a Windows box, so a signed job on PC 2 (run-20261007-172000, woken 18:20 BST) builds both, reads --version off the three exes and posts the installer back through the relay; the smoke (install silent, --version read, the app starts and exits clean) runs on PC 2 by a following signed job and is the gate line. PC 2's agent polled the relay at 18:23 BST while its app, node and miners had been down since 17:27 BST (the Intel driver install), so the jobs reach it; no job on PC 1.

**The Mac:** Igneum-Miner-0.3.21.dmg rebuilt under the build lock at 18:15 BST with c4459193's Mac node pair (igneumd-mac b306baba, igneum-miner-mac deb4d263, the 0.3.20 release's bytes): 490919d9, 44,538,877 bytes, version 0.3.21 build 202610071714. The earlier fb517a11 (96161037's Mac node) never ships.

**Publish reading:** about 19:15 BST on the smoke's green; the staging in a scratch copy with the live floor file and the 0.3.20 hive package unchanged; the apps on the pollers, the Mac first.