self-eval

You can evaluate AI work quality using a two-axis scoring system that rates task ambition and execution independently, then persists scores to a history file to detect inflation and clustering across sessions.

Two-axis scoring rates task ambition and execution quality independently, then combines them via a lookup matrix to provide a balanced evaluation that resists score inflation.

To stop score inflation, the evaluation requires mandatory devil's advocate reasoning, arguing for both higher and lower scores before finalizing, and appends results to a history file to flag clustering.

Yes, the evaluation process appends scores to a persistent history file, analyzing past results to detect score clustering and prompting for re-evaluation if inflation is suspected.

The best way is to require arguments for both higher and lower scores before finalizing an evaluation, combining this devil's advocate reasoning with a two-axis lookup matrix for objective scoring.

No, the evaluation operates entirely in a text-based interface without external tools, relying solely on its internal scripts to manage scoring, devil's advocate reasoning, and history persistence.