fix(perf): require usable repeat references and complete metric observations - #22
Merged
Merged
Conversation
plotarmordev
changed the base branch from
feat/cold-prefill-ladder-15
to
main
September 10, 2026 07:03
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #14. Builds on the descriptive ranges/--reference support from PR #10.
A supplied reference can no longer silently vanish from qualification. Reference-aware changes and drift require complete eligible reference evidence, matched ordered completion counts and complete matching baseline declarations. The output exposes reference model/deployment and declared-match/mismatch/unavailable status; matching declarations do not verify physical restoration or a causal A/B/A experiment. Ineligible/missing evidence stays inspectable with explicit reasons and an ineligible exit, not an automatic two-run fallback.
Reference-aware stream metrics also require all paired lane observations across A/B/A2; a few surviving prefill lanes cannot silently replace the full metric. Other available metrics and the ordinary no-reference path retain their scope. Comparison JSON advances to version 3 for the changed reference eligibility contract. Old execution receipts remain readable and unchanged.
Proposal reconciliation: retained min/max ranges and pooled reference ranges are a coarse descriptive filter. This does not add bootstrap confidence intervals, within_noise significance or the proposed absolute-drift threshold. Those require a separate validated statistical design; observed range separation does not certify significance or equality.
Verified: independent SIX/HELP reviews (both initially blocked partial lane dropout; both approved its fix), exact-head fmt/Clippy/workspace tests/release build; regression failed before the dropout fix and passed afterward; loopback CLI tested valid and undeclared repeat behavior, saved pre-change runs remain readable. No real A/B/A inference run or server restoration verification was performed.