Repository navigation
feat(bench): retain full cohorts in paired task latency - #1032
Merged
Merged
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add an offline scorer and CLI for three declared paired task repetitions. Planned rows and every outcome remain visible; incomplete, unsuccessful or incomparable eligible evidence suppresses whole-cohort completion estimates. Repetitions keep separate median p50, nearest-rank p95, signed deltas and paired ratios. Successful elapsed time beyond the declared timeout/exposure also suppresses metrics; equality is allowed.
Validation: R2 has 70 passing affected tests after three red budget controls, plus a copied deadline-guard assertion mutant. R1's 120 tests and two other mutants remain historical with exact reuse bindings. Scoped lint/format, executed synthetic guide, incomplete-cohort CLI and independent source/correctness/test/evidence/final publication reviews passed. Four frozen sources are preserved through merge master; METHOD and Completion tools link to the contract.
Inputs and hashes are supplied declarations, not authenticated capture or proof of prior freezing/completeness. No actual task latency, provider/native execution, calibrated overhead, reproducibility or W8 acceptance is claimed. Existing metrics/resource APIs and dependencies are unchanged.