How AWE reached each attribution

Follow numbered evidence, assessments and selection decisions. Click a node to inspect its sources and highlight the inputs leading to it.

These are provenance DAGs. Arrows explain the diagnosis process; they do not prove an operation caused a failure. Reasoning numbers differ from original trajectory step IDs.

Observed inputs / artifact evidenceRule inference / policyLLM assessment / arbiterReported decisionUntested check / refuted entryOrigin unavailable / conflicting

Solid arrows: documented evidence and decision inputs. Dashed arrows: a proposed validation check. Typed edge relations appear in the selected node's details. Hidden branches still participate in recorded policy calculations.

Observed action/result links and recovery episodes

Protocol links are kept separate from attribution. A later success on the same action is recovery evidence, not proof that it fixed the task.

Original trajectory excerpt and step references

This is the existing bounded dossier excerpt. Omitted events and unavailable reasoning remain unavailable; no full-trajectory reconstruction is claimed.