Skip to content

Surface agent code in OverallBestEntry for AgentDetailPage#87

Merged
PunchTheDev merged 1 commit into
mainfrom
punch/overall-best-code-link
Jun 4, 2026
Merged

Surface agent code in OverallBestEntry for AgentDetailPage#87
PunchTheDev merged 1 commit into
mainfrom
punch/overall-best-code-link

Conversation

@PunchTheDev
Copy link
Copy Markdown
Owner

Summary

GET /leaderboard/overall and GET /rounds/{id}/leaderboard now include agent_path + commit_hash on every per-spec best record. Closes the flywheel gap on the dashboard's AgentDetailPage — per-problem rows can now deep-link to the exact commit of code that achieved each score.

Motivation

Operator directive (cad-optimizations-todos): "you show the agent that achieved SOTA for each problem and their performance — but NOT their agentic code. The whole point is that the winning agent code gets open-sourced so the next person can fork it and beat it by a decaying margin. Surface the code."

HeroStats (spec detail), SotaHero (homepage), and Leaderboard (per-row) already link to agent code on GitHub. AgentDetailPage was the missing surface — OverallBestEntry carried score + submitted_at but no path/commit, so the per-problem result rows had nothing to link to.

Changes

  • app/models.py: add agent_path: str and commit_hash: str to OverallBestEntry.
  • app/routes/leaderboard.py: SELECT s.agent_path, s.commit_hash and pass through.
  • app/routes/rounds.py: same change in the round-leaderboard path.
  • CHANGELOG.md: 0.15.11 → 0.15.12.

24 existing tests in test_overall_leaderboard.py + test_rounds.py continue to pass — the new fields are populated from columns the fixtures already provide.

@PunchTheDev PunchTheDev merged commit c293d43 into main Jun 4, 2026
1 check passed
@PunchTheDev PunchTheDev deleted the punch/overall-best-code-link branch June 4, 2026 10:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant