From Evidence Ownership to Dynamic Diagnosis in a Multi‑Agent Runtime

Описание и применение

Система CodeFlowMu внедрила динамический диагностический инструмент, который отображает состояние связей доказательств (связанные, отсутствующие, конфликтующие, не применимы) в зависимости от стадии задачи, улучшая прозрачность и точность обзоров.

Задача

Task lifecycle stages (active, review, QA) require different evidence relationships, but existing systems assume a single accountability chain, leading to hidden inconsistencies and inaccurate diagnostics.

Результат

Operators now see explicit evidence states per lifecycle stage, reducing hidden inconsistencies and enabling accurate reviews without altering task state.

Рабочий процесс

• 1. Extract explicit task keys from REPORT writes and ledger entries. 2. Classify relationships: linked (keys match), missing (key absent), conflict (keys disagree). 3. Build evidence graph: TASK→attempt→lease→execution→action evidence→REPORT→REVIEW/EVAL. 4. For each edge, determine state based on current lifecycle stage and role. 5. Render diagnostic snapshot in UI, showing state per edge. 6. Provide actions: copy reconciliation summary, recheck evidence association. 7. Operators use these to verify evidence before advancing task state.

Доказательства

Evidence requirements change with lifecycle stage and role, and the diagnostic changes with the formal facts instead of forcing every edge into linked or missing.

The V2.0.4 task detail states: "This conclusion describes evidence relationships only. It does not mean the task has been delivered or verified successfully."