Skip to content

Fix live pilot resume after shared-ledger copy (#11) - #125

Merged
w4ffl35 merged 1 commit into
masterfrom
issues/w1-2026-10-12
Oct 12, 2026
Merged

w4ffl35 merged 1 commit into
masterfrom
issues/w1-2026-10-12

Conversation

@w4ffl35

@w4ffl35 w4ffl35 commented Oct 12, 2026

Copy link
Copy Markdown
Member

Partial progress on #11 (not closing: the frozen-predictor protocol remains open).

Resume used to restart epoch numbering from the local journal only, colliding with epochs already recorded in a copied shared/global ledger. Epochs now advance past the max recorded in either ledger, in both development and held-out collection, with a regression test covering resume.

Verification: npx tsc --noEmit clean; live-runner suite passes.

collectLiveDevelopment/ Heldout computed the next attempt epoch from the
fresh experiment journal alone, so resuming in a new output directory
after copying global-budget.jsonl (the documented frozen-protocol step)
aborted with 'stale attempt epoch'. The next epoch now advances past the
highest epoch recorded for the run in either the shared global ledger or
the experiment journal.

Adds a regression test seeding a copied ledger with epochs 1-2 and
asserting the resumed attempt starts at epoch 3.
@w4ffl35
w4ffl35 merged commit 7fd1f01 into master Oct 12, 2026
2 checks passed
@w4ffl35
w4ffl35 deleted the issues/w1-2026-10-12 branch October 12, 2026 23:31
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