Status: IN
The system finds problems in itself. The derive-review cycle surfaces errors (13-38% retraction rate), and corrections cascade through the network, improving the foundation for the next derivation round. Each cycle builds on a cleaner network than the last.
Source: repo:beliefs-pi/CLAUDE.md
Depends on (SL): derive-then-review, evidence-retraction-rate