Forge
Forge · Global Leaderboard

The Forge Leaderboard.

Every edition is judged by the same automated multi-agent harness that clones, runs and forensically analyses each build against ground truth. These are the scores only. The full methodology and per-builder breakdowns live on each edition's own results page.
Forge type
Loading…
Edition
Loading…
Rank Builder Score Accuracy
Loading

Fetching the latest standings…

How scoring works

Scoring is objective-first. The largest share of every score is detection accuracy (F1 measured against a held-out ground truth), whether the pipeline actually runs end-to-end, and process integrity (real commits, logs and an honest build trail). Hard integrity gates apply to anything fabricated, hard-coded or plagiarised. Final standings are confirmed on stage on event day.