This paper proposes a compact, inspectable triage layer for autonomous vehicles that captures weak, short-lived edge-case findings and promotes them through a bounded state ladder (V1–V4, R0, Q0). Deliverables include a minimal event schema, deterministic promotion predicates with timing windows, planner action bands per state, and a minimal transition record for V3/V4 transitions. The module is intended as an insertion point between local anomaly emitters and the planner; it is not a perception module, a planner, or a certification framework. The paper provides example thresholds, pseudocode, and a reproducible evaluation plan to measure false escalation, missed early-warning, and planner stability.

Part of the broader Spanda architectural framework repository:https://github.com/putmanmodel/spanda-architectural-framework This bridge paper is a narrow architectural note on early AV edge-case triage. It does not claim a full autonomous-driving stack, safety certification, or deployment readiness.