Skip to content

chore(verification): record the full flagship regeneration as its receipt - #53

Open
nilsonfh wants to merge 1 commit into
mainfrom
chore/record-the-flagship-run-receipt
Open

nilsonfh wants to merge 1 commit into
mainfrom
chore/record-the-flagship-run-receipt

Conversation

@nilsonfh

Copy link
Copy Markdown
Contributor

Follow-up to the topology re-base (#52). The flagship-reproduce receipt claimed the shape-covering subset at 6ea8a92a — 22 stages, 2 of them actually run. The full regeneration has now happened, so the receipt says so.

  • Commit: 9781415d, the re-base merge. The run itself was 6de591bd, squashed into it; the receipt names the merge because the gate requires a commit that exists in this history and leads to HEAD, and a squash discards the branch commit.
  • Run: 25/25 stages completed, manifest completed, 2026-09-15 19:02–22:26Z, cold, in a dedicated worktree.
  • Outcome: regression moved 42 of 94 pins; they were re-based against this run and then verified 94/94 valid. Flagship sweep 157 passed, sense-check valid at 1.0.
  • Evidence: the run manifest's sha256, plus outputs snapshotted outside the repository before any further run.
  • Limit, stated in the receipt: one run proves the pipeline reproduces itself end to end at this commit; it does not attribute each moved pin to each model change, since the four changes moved together.

tools/verification_receipt.py --check passes and tests/test_verification_receipts.py is 18 passed. The protocol no longer reports stale.

🤖 Generated with Claude Code

…eipt

The flagship-reproduce protocol claimed the shape-covering subset at 6ea8a92:
22 stages, 2 actually run. The full ~6 h regeneration CI cannot host has now
been run, cold, from committed code, and its numbers are what the study pins.

  commit   9781415 (the topology re-base merge; the branch commit that ran,
           6de591bd, was squashed into it)
  run      25/25 stages completed, manifest status completed,
           2026-09-15 19:02 to 22:26Z
  outcome  regression moved 42 of 94 pins, re-based against this run and then
           verified 94/94 valid; flagship sweep 157 passed; sense-check valid
  evidence run manifest sha256 3219eca751932e6f..., outputs snapshotted outside
           the repository before any further run

The limit is stated in the receipt rather than left to a reader: one run proves
the pipeline reproduces itself end to end at this commit. It does not attribute
each moved pin to each model change, because the re-base's four changes moved
together.

tools/verification_receipt.py --check and tests/test_verification_receipts.py
pass, and the protocol no longer reports stale.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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