Observed arrival · 2026-10-28
PROBE, the proactive bug-discovery benchmark
An open benchmark compares coding agents’ whole-repository bug audits with 2,457 bugs maintainers fixed across 20 repositories.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers and developers evaluating coding-agent bug audits
- Worth noticing
- The reference set counts bugs maintainers fixed during the 18 months after agent audits; unmatched reports may still be valid.
Field notes
The page groups 20 repositories across Python (10), TypeScript (6), and JavaScript (4), with reference-bug counts shown for projects including sqlglot (512), prefect (429), and Svelte (248). Its leaderboard reports F1, recall, precision, finding totals, and per-run API cost; users can sort columns, filter by language, inspect rows, and download a CSV. It also notes that unmatched findings may still be valid.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue