Your free trial has ended.
Try the live demo or log in .
By: ValentinHirschi <valentin.hirschi@gmail.com>
Expose and diagnose effective finite replay budgets Keep existing aggregate minima and defaults, add explicit application/pending/coalescing controls and a numeric preflight, and preserve typed refusal details in live and cold replay. Eight focused native groups and the full 58-query four-loop control pass; record repeated local rule-work improvements without a full-campaign speed claim. Co-authored-by: Codex <noreply@openai.com>