Skip to content

Promote matrix_multiplication into the governed night loop - #8

Merged
ucsandman merged 2 commits into
masterfrom
devin/1789432844-matmul-night-slot
Sep 15, 2026
Merged

ucsandman merged 2 commits into
masterfrom
devin/1789432844-matmul-night-slot

Conversation

@devin-ai-integration

@devin-ai-integration devin-ai-integration Bot commented Sep 15, 2026 •

Copy link
Copy Markdown
Contributor

What changed

Matrix multiplication now has a paired research slot in the governed night, with the shared allowance, routing journal, Docker workers and local evidence. It runs after the counterbalanced trial pair and before power-grid validation. The default plan is 540 minutes and 105 accounting units. These units measure subscription usage, not cash spending.

Review caught and fixed four integration problems:

  • Docker did not admit the plugin. Workers now receive its trusted exact verifier through the existing allowlist.
  • The median across three matrix sizes rejected a gain on just one size. Matrix multiplication now accepts a replicated per-target gain only when no matched evaluation regresses. Other plugins keep their existing median policy.
  • Dashboard limits still rejected the new 105-unit default. The form and API now match the scheduler's 130-unit ceiling.
  • The installed 22:00 start could not fit nine hours before the 06:00 cutoff. The installer now prepares a 21:00 start and a 9h15m task limit. The catch-up window starts at 20:50. Morning jobs keep their existing times.

Confirmation uses the same targets with fresh seeds and is labeled repeatability evidence. Local incumbent advancement is separate from publication. The existing all-target release gate cannot mark matrix multiplication publishable because its n=2 calibration is already optimal. Individual record publication still needs a separate reviewed path.

Also corrected partial-run scheduling history, stale schedule documentation and historical activation claims. Based on merged PR #7.

Verification

  • python scripts/check.py: 268 passed, 5 skipped; Ruff passed; 76 Python files compiled.
  • Real Docker evaluation of all three incumbent targets: exact verification passed at ranks 7, 26 and 49. The governed evaluation-only command completed 3 evaluations with zero failures, zero model calls and zero allowance charged.
  • Dashboard: 6 allowance API cases passed; actual form saves and reloads at 105 and 130 succeeded; 131 was rejected. Browser route check passed with no console or network errors.
  • Installer preview showed 21:00 / 9h15m and unchanged morning times. No task registration was applied.

Activation

Merging this PR does not change Windows task registration. Existing 22:00 installations still need a separately approved update to provide the full nine-hour window. The installer exports rollback XML before applying changes.

- night.json gains a fixed-provider (paired) matmul research slot, 90 min,
  12+3 units; night allowance 105 units over 540 minutes. The slot is outside
  the counterbalanced trial and orders after the trial pair by information
  gain, before pglib validation.
- evaluation.build_manifest supports CONFIRMATION_ON_DEVELOPMENT: confirmation
  re-runs the same development targets under fresh seeds, classified
  same_target_fresh_seed_replication; the manifest now reports concealed
  targets separately from confirmation targets.
- matrix_multiplication.prompt_for_targets keeps the interface contract,
  strategy notes and honest framing while dropping lines that name withheld
  targets.
- Docs updated: README slot table and allowance, OPERATIONS, DECISIONS
  (2026-09-14 entry), RESEARCH-IMPLEMENTATION, CHANGELOG.

Co-Authored-By: Wes Sander <sandman.uc@gmail.com>
@devin-ai-integration

Copy link
Copy Markdown
Contributor Author

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

  • Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
  • Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

  • Disable automatic comment, CI, and merge conflict monitoring

@ucsandman
ucsandman changed the base branch from devin/1789431760-prompt-context-schedule to master September 15, 2026 01:11
@ucsandman
ucsandman merged commit ce97256 into master Sep 15, 2026
4 checks passed
@ucsandman
ucsandman deleted the devin/1789432844-matmul-night-slot branch September 15, 2026 02:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant