Diagnose agent evaluation failures

Wave Analysis

Root-cause analysis across multi-scenario runs with dedup and regression detection.

• · scenario list: {id, distilled, raw, def, meta}
• · change context string

• · findings array with evidence + root causes
• · ranked findings sorted by impact

• · agent-orchestration-engine
• · parallel-execution-runtime
• · phase-tracking-system
• · logging-framework

• · workflow must export meta={name, description, phases}
• · args may be JSON-stringified; parse defensively
• · multi-phase execution context available

• · agent orchestration failure or timeout

• · Confidence levels distinguish verified vs inferred claims