learning.loop

It gets smarter every merge.

Proof the system improves — review accuracy from real PR outcomes, eval-score trend, self-learned patterns, and AutoFix self-verification.

← AutoFix
the moat compounds
%

AutoFix self-verification rate. Every merged fix mines a rule and a solved case, so the next bug is faster, safer, and eventually one-click.

0 regressions
not enough data yet
self-verification rate · trailing weeks
review accuracy
%
0 PR outcomes recorded
latest eval score
%
harness pass rate
patterns learned
0
0 confirmed catches
autofix verified
%
0 runs · red→green

AutoFix self-verification / week

fill = verified (red→green) share of runs
No data in range yet.

Review accuracy / week

fill = clean share of recorded PR outcomes
No data in range yet.

Eval-run leaderboard

harness
runscorepass/failwhen

Patterns the system learned

0 active clusters