fix: resumable parity reference decode (the five silent gate deaths) - #82
Merged
Conversation
Five consecutive ara-diac2 gate attempts died mid-reference-decode (39min-2.5h, variable, traceback-less container kills — localized by the stage log: every attempt logged start, none logged reference-decode done). The reference now checkpoints per-input to reference_progress.jsonl on the volume and resumes on relaunch — the same pattern the r-series evals have used since r5. Kills now cost only the undecoded tail.
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
The ara-diac2 parity gate died five consecutive times mid-reference-decode — variable lifetimes (39min–2.5h), no traceback, container-level kills. The stage log (PR #70's observability) localized it precisely: every attempt logged
start, none ever loggedreference-decode done.The fix is the pattern the r-series evals have used since r5: resumable progress.
reference_decodenow checkpoints per-input results toreference_progress.jsonlon the volume and resumes from the saved prefix on relaunch. Container kills now cost only the undecoded tail; the retry loop converges instead of restarting from zero each time.Parity tests green, ruff clean.