Whether every step hit or healed (no unrecovered divergence).
Steps that diverged (replay+heal failed) — the UI-drift signal.
Whether a divergence forced an in-attempt fall back to the LLM tier.
Steps that self-healed (recorded target drifted, re-resolved).
Number of script steps replayed.
The replay sub-tier the run dispatched into.
Compact per-attempt replay outcome a tiered driver may surface. Mirrors the engine's replay counts so a report's replay-health panel and the "is replay lying to us?" check can be computed without the full step list.