fix: rank leaderboard header cards by the same averaged data as the charts - #148
Conversation
The header leaderboard computed its own aggregate — an unweighted mean over every variation x fixture cell with DNF results silently imputed as the slowest successful time — and sorted the default (average) view by wins first. Card order could contradict the times printed on the cards and the averaged chart data. Reuse calculateAverageVariationData (the same calculation behind the charts' average data, which already filters out DNF results) to build the per-fixture values, and sort by average time in every view so rank always matches the displayed value. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Matches the vlt.io/vltpkg convention — CLAUDE.md and .claude/ skills stay local and untracked. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ruyadorno
left a comment
ruyadorno
left a comment
There was a problem hiding this comment.
I believe there was a confusion in what "chart data" should have been used as a reference here, the chart that contains the correct calculations is the "Performance Over Time" chart and it does indeed impute the slowest time as the time to be used for a DNF run.
Please change the implementation so that the leaderboard header cards are ordered using that same logic as the "Performance Over Time" chart, the numbers displayed on the leaderboard header cards were correct and the changes to that should be reverted here.
Address review: the leaderboard card numbers were correct (DNF runs imputed as the slowest successful time, same as the Performance Over Time chart data) and are reverted to the original calculation. Only the ordering changes: cards are ranked by that average time in every view, with wins as a tiebreaker, instead of by wins. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Thanks for the catch — you're right, the "Performance Over Time" chart data does impute DNF runs as the slowest time, so the original card values were already consistent with it. I've reverted the value calculation to the original logic (af7a386); the only remaining change is the ordering: cards are now ranked by that same average time in every view, with wins as a tiebreaker. PR description updated to match. |
3 participants
Problem
In the default "average" view, the leaderboard header cards were sorted by
wins(counted per raw variation × fixture cell) rather than by the average time printed on the cards — so card #1 could display a slower time than card #2, contradicting the card values and the "Performance Over Time" chart.Fix
The card values are computed exactly as before: per raw variation × fixture cell, with DNF runs imputed as the slowest successful time for that fixture — the same DNF handling the "Performance Over Time" chart data uses. The only change is the ordering: cards are now ranked by that average time (lower is better) in every view, with wins kept as a tiebreaker only.
Route/data-source selection (
usePerPackageData, registry vs task-runner vs package-manager) and the fixture filter behave as before.Also adds
.claudeto.gitignore, matching the vlt.io/vltpkg convention of keeping local agent config untracked.Testing
npm run build(tsc + vite) passesnpm run lintpasses🤖 Generated with Claude Code