Fix duplicate leaderboard PRs; add reset_results knob#34
Closed
isPANN wants to merge 1 commit into
Closed
Conversation
…e title Each scoring run opened a fresh bot/leaderboard-<run_id> PR, so triggering the workflow twice produced duplicate identical PRs (e.g. #32/#33) with no dedup. Use a single rolling branch `bot/leaderboard`, force-updated every run, and create the PR only if one isn't already open (else update its title) — so there is at most one open leaderboard PR ever. Also add a `reset_results` workflow_dispatch input that wipes R2 results/ before scoring, to clear stale scored files (e.g. a pre-test-marker entry that keeps resurfacing on the board), and fix the "1 models/bugs" pluralization. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018xw6L5yUrz33UrQpanbwnH
Collaborator
Author
isPANN
added a commit
that referenced
this pull request
Jul 6, 2026
The public leaderboard was a single site/results.json updated by one aggregate PR, so all models were coupled — you couldn't accept, hold, or revert one model's result without the whole board, and triggering the scorer twice opened duplicate PRs (no dedup). Make each submission its own reviewable unit: - Source of truth is now one file per submission, site/results/<slug>.json, where slug = <model>--<UTC-time>--<short-id>, derived deterministically from the scored file (backend_score.board_slug/board_entry/write_board_entries). Test submissions produce no public file. - score-from-r2.yml opens ONE PR per new/changed entry on branch bot/leaderboard/<slug>. The slug is stable, so re-scoring the same submission force-updates the SAME branch/PR (idempotent — no duplicates); a different submission is its own independently mergeable/revertible PR. - site/results.json becomes a generated artifact: publish-on-merge aggregates site/results/*.json (best run per model) via `backend_score --build-board`, guards each entry + the built board, and deploys. It is gitignored; `make board` / `make serve` build it locally. - check_aggregate accepts a single-entry object (per-submission file) as well as the list, and allows the timestamp/submission_id provenance tags. - Also adds a reset_results workflow input (supersedes PR #34) and fixes the root-anchored /results/ gitignore so it no longer shadows site/results/. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018xw6L5yUrz33UrQpanbwnH
isPANN
added a commit
that referenced
this pull request
Jul 6, 2026
The public leaderboard was a single site/results.json updated by one aggregate PR, so all models were coupled — you couldn't accept, hold, or revert one model's result without the whole board, and triggering the scorer twice opened duplicate PRs (no dedup). Make each submission its own reviewable unit: - Source of truth is now one file per submission, site/results/<slug>.json, where slug = <model>--<UTC-time>--<short-id>, derived deterministically from the scored file (backend_score.board_slug/board_entry/write_board_entries). Test submissions produce no public file. - score-from-r2.yml opens ONE PR per new/changed entry on branch bot/leaderboard/<slug>. The slug is stable, so re-scoring the same submission force-updates the SAME branch/PR (idempotent — no duplicates); a different submission is its own independently mergeable/revertible PR. - site/results.json becomes a generated artifact: publish-on-merge aggregates site/results/*.json (best run per model) via `backend_score --build-board`, guards each entry + the built board, and deploys. It is gitignored; `make board` / `make serve` build it locally. - check_aggregate accepts a single-entry object (per-submission file) as well as the list, and allows the timestamp/submission_id provenance tags. - Also adds a reset_results workflow input (supersedes PR #34) and fixes the root-anchored /results/ gitignore so it no longer shadows site/results/. Claude-Session: https://claude.ai/code/session_018xw6L5yUrz33UrQpanbwnH Co-authored-by: Claude Opus 4.8 (1M context) <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.
Problem (spotted while testing): triggering
score-from-r2twice opened two identical leaderboard PRs (#32/#33) — each run used a freshbot/leaderboard-<run_id>branch, so there was no dedup.Fixes:
bot/leaderboardbranch, force-updated each run; create the PR only if none is open, else update its title. At most one open leaderboard PR ever.reset_resultsinput — a manualworkflow_dispatchboolean that wipes R2results/before scoring, to clear stale scored files (e.g. a pre---test-marker entry that keeps resurfacing on the board).1 model, not1 models).Note: the two
--testsubmissions were correctly excluded from the board; the lingeringclaude-sonnet-4-6: 1was that stale pre-marker scored file in R2 —reset_resultsclears it.🤖 Generated with Claude Code
https://claude.ai/code/session_018xw6L5yUrz33UrQpanbwnH