# CAIN-42 E21 — Replay & State-Integrity Report

## Deterministic distributed replay
- 1,000-step chain: `replay_transitions` recomputes the root from each committed diff; deterministic.
- Differential replay: 4 independent replays agree on exactly one final root.
- Evidence: `artifacts/e21/E21_REPLAY_RESULTS.json`.

## State integrity
- Each transition binds `parent_state_root`; a mismatch fails at the exact step (`failure_at`).
- A declared `proposed_state_root` that disagrees with the recomputation is a determinism failure.
- Invariants I28 (continuity), I29 (parent binding), I31 (deterministic evaluation), I42 (historical
  immutability) enforce this; 24/24 pass (`E21_INVARIANTS.json`).
- Divergence detector returns the earliest divergent index (`E21_DIVERGENCE_RESULTS.json`).

## Not done
- Multi-host replay differential (four real replicas) — NOT_MEASURED.
- Long-horizon soak (1h/6h/24h/72h/7d) — NOT_MEASURED.
