Changelog
Site updates
Leaderboards published, method changes, corrected readings, and changes to the site itself. A method change is recorded here before the numbers move.
Leaderboard added
Patch leaderboard
Patch leaderboard published. 18 systems added.
data unchanged
Leaderboard added
Exploit leaderboard
Exploit leaderboard published. 3 systems added.
data unchanged
Data
Every leaderboard row names the system's harness, model and reasoning effort.
Data
Attack leaderboards show triggered and exploited separately.
Data
Defend leaderboards show fixed and any-credit separately.
Data
Every column is sortable.
Site
The Unpatched leaderboard moved to /benchmarks/unpatched; the earlier /benchmarks/zero-day address redirects there.
Data
Unpatched and Cryptography leaderboards published, with two research notes.
Data
Attack versus defend published as a paired view of the Vulnerability repair data.
Method changed
Patch leaderboard
Intervals recomputed with codebases as the unit; systems whose ranges overlap share a rank.
data version 052fee2d42becd04
Reading corrected
Patch leaderboard
Grade mapping corrected for passes with no failure class; counts and rates unchanged.
data unchanged
Data
Patch and Exploit leaderboards published; Attack versus defend published.
Site
Site rebuilt around the leaderboards, with the method on its own page.
Site
llms.txt generated from the research notes.
Site
Research note published: a model graded before it had a public name.
Site
Model pages, pairwise comparisons, an economics page and the first research notes published.