Catch regressions that change run-to-run behaviour without moving the… - #4883
Merged
Conversation
… median The compare tab's rank test answers "did runs get generally slower" and is blind to a metric whose median holds while the runs split into two groups, for example a race that sends a third of the iterations down a slow path. The distribution strip draws that pattern but nothing tested it, so the row read "no change" at p around 0.5 while the spread visibly exploded. Every compared metric now also runs a two-sample Kolmogorov-Smirnov test, which compares the whole distributions. It only surfaces when the rank test saw nothing (K-S fires on plain location shifts too, and those rows are already flagged) and when the one-sided likely-faster call doesn't own the story: the row goes amber with "spread changed", the KS p-value and the change in the middle-80% span, and the verdict card lists it separately as possible intermittent behaviour rather than a regression. At 20-ish iterations this catches clear variance growth and bimodality, not subtle shape drift, and the docs say so. The new statisticalTestKS field is additive; the graphite export is untouched. Co-authored-by: Claude Fable 5 noreply@anthropic.com
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
… median
The compare tab's rank test answers "did runs get generally slower" and is blind to a metric whose median holds while the runs split into two groups, for example a race that sends a third of the iterations down a slow path. The distribution strip draws that pattern but nothing tested it, so the row read "no change" at p around 0.5 while the spread visibly exploded.
Every compared metric now also runs a two-sample Kolmogorov-Smirnov test, which compares the whole distributions. It only surfaces when the rank test saw nothing (K-S fires on plain location shifts too, and those rows are already flagged) and when the one-sided likely-faster call doesn't own the story: the row goes amber with "spread changed", the KS p-value and the change in the middle-80% span, and the verdict card lists it separately as possible intermittent behaviour rather than a regression. At 20-ish iterations this catches clear variance growth and bimodality, not subtle shape drift, and the docs say so. The new statisticalTestKS field is additive; the graphite export is untouched.
Co-authored-by: Claude Fable 5 noreply@anthropic.com