zicato-triage-stuck-loop

Rejected candidates and flat scalar improvements indicate a stuck evolution loop caused by degenerate evaluation signals, overly difficult boards, restrictive Forbidden sets, or repetitive proposer behavior. Diagnosing these specific conditions determines the correct contract or steering change to restore progress.

Diagnosing weak evaluation signals involves running read-only zicato health checks and SQLite index inspections to identify flat drift, dead judges, or non-differentiating entries before adjusting the board or scoring contract.

The best way to troubleshoot repeated rejections is to inspect tournament, loss-profile, experiment, and patch history with safe read-only SQL queries, then determine whether to revise the board, relax an over-demanding goal, or adjust the Forbidden set.

You can inspect SQLite index diagnostics and run read-only zicato health checks without spending LLM budget, allowing you to review board, scoring, and proposer-brief contracts safely before starting a new evolution round.

You should adjust the Forbidden set or re-steer the proposer brief when health checks reveal the needed mutation is forbidden or the proposer is repeating an ineffective idea, preventing the evolution loop from distinguishing candidates.

Zicato triage works for flat drift and dead judges by applying read-only health findings to decision-tree troubleshooting, distinguishing whether the board cannot distinguish candidates or the task is simply too hard.