igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity
978 commits 218 branches 0 tags 1.1 GiB
Commit graph

978 commits

This branch
This branch
All branches
Author SHA1 Message Date
igneum-labs
0b48c028f3 Ember helper for 0.3.15, the 0.3.14 window's five causes: (B, C) the helper writes a heartbeat every poll, the engine starts the task and waits for a fresh heartbeat before writing any command, the acknowledgement is the helper's own log line; (D) a measure-only card is never set (ember::may_set); (E) an explicit Tune now clears the failure back-off and a held row says when the retry comes (ember::retry_note); tests. The shipped-app window playbook's gate at 0.3.15; relay/playbooks/ember-tune-pc1.ps1 deleted (superseded); the post-mortem table and the 0.3.15 next-cut table in the plan; bench log 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 18:39:00 +00:00
igneum-labs
2b244d732e Ember: a reads-only log grep playbook (the app's own tune lines, the helper log, cmd.txt) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 18:33:19 +00:00
igneum-labs
0659fab5f2 Ember helper, two bugs from PC 1 on 6 October 2026: (A) a helper acts only on lines added after its start, so a stale quit or remove never kills a later start (commands_after + test); (B) the engine starts the task again before every task-path step and the acknowledgement is the helper's own log line for the sequence (up to 15 s), a missing line fails the step with that reason instead of a blind 4-second ack. Playbooks read the app log from <data root>\logs, where it lives 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 18:31:38 +00:00
igneum-labs
14dc093549 Ember shipped-app window playbook: the version gate on the first line (0.3.14 or later, the climb field present, else abort before anything is asked), the prompt count from the app's log over the window, the climb against run 6's ladder floor per NVIDIA card 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 17:00:30 +00:00
igneum-labs
59631f07c7 Ember: every task command goes to the helper's fixed folder (platform::fixed_data_root, powertask::helper_dir, engine helper_cmd_dir): a scratch-root engine wrote under its own root, where the task never reads; test. The shipped-app window playbook: the installed app tunes itself through the task (AMD ladder, then the NVIDIA climbs), asked and watched over its API, never stopped 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:59:33 +00:00
igneum-labs
db0909f72d Ember: a baseline (measure-only) result's row reads Measured, not Tuned; the event line no longer doubles the words; test; next-cut table row 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:56:24 +00:00
igneum-labs
a787e9e682 Ember: the re-point and the no-prompt proof in the plan and the bench log 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:55:03 +00:00
igneum-labs
ac6731b2dd Ember proof playbook (reads only): the installed app's power cap lines, the helper log, the task's action, nvidia-smi, the consent.exe count, the cards' state 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:52:12 +00:00
igneum-labs
28c4e81039 Ember re-point playbook: the helper.log line is read as an array element, not the last character of a string (the run printed helper_log=r) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:51:16 +00:00
igneum-labs
cd33f5d9f0 Ember plan: the ADLX offset-not-watts rule in the AMD knob row; the 0.3.14 next-cut table (57de311, 12e045b, the Ember 2 UI and playbook commits) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:45:39 +00:00
igneum-labs
887bff36aa Ember run 6 closed: the 4070 table and the 9070 XT abort in the plan and the bench log, the 4070 row on the fleet page, the playbook prefers ember-kit-7 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:44:36 +00:00
igneum-labs
57de3112b8 Ember: a power step on a card with no limit readback (AMD through ADLX reports an offset, never watts) is applied on the acknowledgement; run 6 aborted the 9070 XT's ladder at its first step on "card reports 0 W"; test 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:43:26 +00:00
igneum-labs
12e045b64e Ember run 6: the helper registered on the scratch copy, so task_exe prefers the installed exe for every path that is not itself an install candidate, and the helper gains a reregister verb (no path argument; re-points the task at the install folder it finds itself, logs the action read back); the clock ladder and floor go to 45% with the 1% rate tolerance as the guard; a chosen point on the floor says floor, not optimum; the 5090 row (84 W saved for 0.15% of rate) on the fleet page; the re-point and no-prompt proof playbook; run 6 notes, threat note, bench log 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:35:03 +00:00
igneum-labs
f20cc03b98 Ember playbook: the AMD watts source. The script samples the installed app's /api/state power_w for the AMD card every 10 s (a GET), counts the engine's own AMD watts, prints a fallback ladder (median per step, with the step's MH/s and MH/W) when the engine's AMD progress carries no watts, and names the source it used 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 16:00:09 +00:00
igneum-labs
07f04243d5 Ember 2: the plan's section 6b (the clocks, the hill-climb, the goal, the price, the curve, the fleet prior, what is owed); Settings: the goal, the electricity price, the climb switch; the card detail: £ a day for the tuned point and the measured curve; UI test 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:56:22 +00:00
igneum-labs
eda64d599f ember-tune-pc1.ps1: ember-kit-6 (the quote fix, the self-registering elevated engine) preferred when present 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:52:37 +00:00
igneum-labs
229ca3f091 Ember 2: the climb's tests (a synthetic memory-bound card converges in five probes; a refused probe backs the knob off for good; each goal picks its point; the £/day formula) and the fleet prior carrying the memory clock (climb records aggregate) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:50:34 +00:00
igneum-labs
0bf27b1d78 state: tune_period_s on the settings state (the manifest's period), so the UI derives the next tune from sweep_at 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:48:44 +00:00
igneum-labs
0ef80a99bd The one approved Power control step on PC 1 failed with exit 1 (15:45:21Z, no cap applied, no helper registered): cmd.exe strips the first and last quote of a /c line that starts with one and holds more than two (two NVIDIA cards, or a cap plus the registration script); both launchers now wrap the line in one outer pair (platform::cmd_c_args, host.cpp runElevated), unit-tested; Ember 2: the memory clock set through nvidia-smi -lmc/-rmc (direct and through the helper's lmc/rmc verbs), the goal, the electricity price and the hill-climb switch in Settings and /api/tune/goal, the chosen point's memory clock and the measured curve on the card 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:48:14 +00:00
igneum-labs
c5e5de6752 Run 5 (15:28 to 15:39Z): the 5090's five power rows in the bench log (the cap does not bind: 0.41 MH/W flat at 310 W); the playbook's watchdog killed the live three-card tune at the first clock step, so: idle = three consecutive status lines reading 0.00 MH/s and 300 s, never a missing match; the after snapshot and the engine-log dump on every exit; an elevated tune engine registers the Igneum Power Helper itself before its first step (the --sweep engine skipped the cap path where the registration lived); Ember 2 groundwork: the memory-clock knob in Point and Limits, the goal, the hill-climb plan (not yet wired to the engine) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:44:06 +00:00
igneum-labs
7ed0d75160 A card never shows 0 MH/s without a reason word (the project lead, 6 October 2026, watching PC 1 during the table run): the Mine row reads 'tuning: step k of n · measuring W W' with the live rate, 'held for a remote job: <title>' while a job holds the miners, and the Tuned line at the end; the big button reads 'Tuning, mining again in about N min'; the strip carries one tune line; a measurement engine prints a TUNE progress line every 10 s and the playbook forwards it (and TUNE chosen as done) to the installed app's POST /api/tune-progress, a post of state, never quit, pause or resume; the rule in ember-tune.md 6a; UI tests for the three states 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:36:07 +00:00
igneum-labs
7bc0e1ce1a powertask: a kit run from a jobs folder registers the INSTALLED exe as the task's action (the 15:28Z run on PC 1 registered jobs\ember-kit-5\igneum-app-ember.exe); the playbook uses a kit only on an app that predates Ember Tune 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:30:42 +00:00
igneum-labs
6099f37ed5 ember: a registered Igneum Power Helper task counts as permission for the tune (an unelevated measurement engine, Power control forced off, still sets NVIDIA limits through the task with no prompt) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:11:37 +00:00
igneum-labs
731d3e70c3 ember-tune-pc1.ps1: budget 55 min (three cards: 5090 9 steps, 4070 10 steps, 9070 XT 4 steps, 75 s each) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:10:53 +00:00
igneum-labs
452d74c3c2 playbook-quit-check: the ruling of 6 October 2026 15:30 UTC (a job script never pauses or resumes the installed app's miners, not even as a fallback; --stop-miners is the way): the two agg-cost allow entries dropped, so the gate names tools/proving-v1/pc2-agg-cost.ps1 and pc2-agg-cost-restore.ps1 until they move to --stop-miners 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:05:37 +00:00
igneum-labs
e9e1042fc9 playbook-quit-check: pc2-agg-cost.ps1 and pc2-agg-cost-restore.ps1 allowed pending the rule owner's word (they pause and resume the installed app as the fallback of a card switch) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:04:35 +00:00
igneum-labs
4af1d66917 Merge master 017c7db into ember-tune (0.3.12's version files and the proving-v1 app merge; my own 27c2db6 came back as 4965220) 2026-10-06 15:03:37 +00:00
igneum-labs
caa24cd5eb bench log: dry run 3 passed (the first measurement engine on PC 1 that mined: 5090 127.31 MH/s at 316.5 W, 4070 28.68 MH/s at 102.7 W), the folder-lock finding measured on PC 1 and its consequence 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 15:03:18 +00:00
igneum-labs
0cba03027e lock_permissions: the folder grant is user:(OI)(CI)F WITHOUT /T (measured on PC 1, collect ember-acl-2: /T re-applies /inheritance:r to each file after the propagation and an (OI)(CI) entry on a file is inherit-only, so the copied settings still read as nothing); the playbook prefers ember-kit-5 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 14:50:49 +00:00
igneum-labs
0da0b65180 ember-tune-pc1.ps1: prefer ember-kit-4 (the ACL fix) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 14:41:31 +00:00
igneum-labs
add025fb25 Why every measurement engine on PC 1 ran on defaults: lock_permissions cut the app folder's inheritance with a non-inheritable user:F, which left files COPIED in before the start (settings.json, machine-id, wallet.json) with no ACE at all, unreadable by their owner; the folder grant is now user:(OI)(CI)F with /T (unit test on the argument shape); a --sweep engine never asks for the firewall rule; the playbook copies the firewall flag; the settings fixture test 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 14:39:18 +00:00
igneum-labs
f225ccf842 run 4's cause: the playbook's PowerShell JSON round trip rewrote the copied settings (big integers as doubles), the engine read the file as defaults (no payout address, every card off, 96 old remote jobs run in the scratch root). Fix: the copies are verbatim and a --sweep engine applies Settings::for_measurement in memory (remote jobs, updates, proving and Power control off, not paused, tune on, every card due and unpinned); the CI check fails any playbook that rewrites settings.json through ConvertTo-Json; unit test 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 14:21:06 +00:00
igneum-labs
eb3d9e51f3 ember-tune-pc1.ps1: the watchdog's exits dump the engine's log tail and stderr first (run 4 left no engine line) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 14:18:29 +00:00
igneum-labs
280631fbc6 tools/ci/playbook-quit-check.sh: the standing rule of 5 October 2026 23:05 UTC as a gate (a playbook that reads the installed app's URL file and sends quit, pause or resume fails; the installer's own stop step is the one allowed sender; self-test on a bad and a good case); shard-test.ps1 loses its api/quit to the installed app; ember-tune-pc1.ps1's refusal guard reworded 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:52:08 +00:00
igneum-labs
5888eea534 One administrator approval, ever (the project lead, 6 October 2026 11:50 UTC): the first Power control approval also registers the per-user scheduled task 'Igneum Power Helper' (RunLevel Highest, no trigger, action = the app's own exe --power-helper); every later cap and every tune's helper starts the task and writes the command file, no prompt, across restarts, updates and reboots; Power control off sends remove and the task unregisters itself; the helper runs only fixed verbs with digit-only nvidia-smi arguments (threat note in ember-tune.md 7a); 4 tests 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:50:49 +00:00
igneum-labs
017c7db517 Igneum Miner 0.3.12: the fresh-record rule switch (proving v1), the segment-aligned prover, Ember Tune, the GPU list in performance order 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:48:10 +00:00
igneum-labs
214d0be22a release 0.3.12: the plan's sections deduplicated and in order 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:47:55 +00:00
igneum-labs
201d38a49c release 0.3.12: PC 1's minute at 0 MH/s after the switch (the C43 class); sections in order 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:47:32 +00:00
igneum-labs
32febb5f56 ember-tune-pc1.ps1: a fresh scratch root per run (run 3 at 11:45Z: the folder an earlier elevated engine had locked refused the settings copy, the engine exited in 6 s on stale files), the installed engine preferred from 0.3.12 on, the address guard hard 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:47:08 +00:00
igneum-labs
0be6575cfd release 0.3.12: the rollout (the re-pin to 198,000, the two publishes, per-machine times, the sweep, the Mac pause on the project lead's order) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:46:55 +00:00
igneum-labs
be0937bc41 ember-tune-pc1.ps1: budget 45 min (two NVIDIA cards, 19 steps) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:44:58 +00:00
igneum-labs
9234ef4be1 release 0.3.12: merge master fc4d82f (the morning summary, docs only) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:28:27 +00:00
igneum-labs
54ec7eb66e packaging/mac/packaged-config.sh: proving_v1_fresh_rule_daa re-pinned to 198000 (the floor read 10,418 at 11:20Z, tip 181,582) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 11:21:33 +00:00
igneum-labs
fc4d82f445 morning summary: 12 GB mine-and-prove measured as a core-only prover 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 09:15:27 +00:00
igneum-labs
4fbf9a1f62 release 0.3.12: the PC 2 job's place in the queue (one second before floor-build-6) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 09:14:21 +00:00
igneum-labs
c7d0927e43 release 0.3.12: the CI verdict (37440559793 on 01db0dd), the gate stands 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 09:08:42 +00:00
igneum-labs
01db0dd726 release 0.3.12: merge master 5334b4a (the morning summary, docs only) 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 09:05:27 +00:00
igneum-labs
6afd307c29 pc2-agg-cost.ps1: pkill -f sp1-gpu-server at the end (the CI root-socket check); release 0.3.12: the PC 2 job, the Windows artefacts, the CI rows 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 09:04:33 +00:00
igneum-labs
8ce243a9ff release 0.3.12: the Windows exes, the second inputs push, the consensus-core fast-time finding 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 08:50:13 +00:00
igneum-labs
b3bb4c9311 Windows payload inputs: node 83089544 (push-inputs.sh, the Mac cross-build, the 0.3.11 workers, the AMD telemetry helper, signed); release 0.3.12 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-10-06 08:49:08 +00:00