learning-systems

Implicit feedback scoring converts task outcomes—success rates, error counts, retry attempts, and execution duration—into learning signals that guide pattern maturity progression. Use swarm_record_outcome() to log bead results; the system weights recent outcomes more heavily via confidence decay to prioritize current performance over historical patterns.

Confidence decay applies a half-life based mechanism that diminishes the influence of older task outcomes, ensuring recent results carry more weight in pattern evaluation. This prevents stale data from blocking pattern promotion or deprecation and keeps learning loops responsive to current swarm performance.

Pattern maturity transitions are driven by aggregated implicit feedback scores across subtasks. The system monitors outcome recordings—duration, errors, retries, success flags—and automatically evaluates when patterns meet thresholds for promotion to production use or demotion when performance declines.

Yes. Record outcomes with files_touched and strategy metadata to identify which patterns and file-based strategies underperform. Implicit feedback signals reveal error patterns, retry clusters, and duration outliers, enabling targeted debugging of pattern decomposition quality.

Capture bead_id, duration_ms, error_count, retry_count, success flag, files_touched, and strategy to enable comprehensive implicit feedback scoring. Complete outcome recording ensures confidence decay, pattern maturity transitions, and anti-pattern inversion work accurately across your swarm learning loops.