reflexion

Root-cause extraction from iteration feedback is handled by parsing ITERATE verdicts, categorizing the feedback items, and storing structured reflections in forged.db. This identifies underlying issues and generates guidance to prevent repeating the same failures in future cycles.

Generating retry guidance after a failed iteration roundtable requires analyzing the stored reflections and detected patterns from previous cycles. The system parses the feedback, extracts root causes, and outputs precise retry guidance to correct the specific failures in the next attempt.

Pattern detection works by parsing and categorizing multiple feedback items across iteration cycles, then checking stored reflections in forged.db for recurring root causes. When repeated patterns are identified, the system triggers alerts to highlight systemic issues.

Issues during feedback processing should be escalated after three consecutive iteration cycles fail to resolve the problem. The system includes guardrails that automatically escalate when retry guidance and root-cause analysis fail to break the cycle.

Storing iteration reflections and tracking their status across features and stages is supported by persisting the categorized root causes to forged.db. The system applies this logic across roundtable iterations, maintaining status checks and pattern alerts for ongoing learning.