Crash-recovery assessment gates

Score each gate 0 missing/incorrect, 1 partial or prompted, 2 independently correct with evidence, 3 fresh-case transfer with limitations. Every gate needs 2; no aggregate score waives a failed gate.

• FS1 — State separation: trace volatile log/data, durable log/data and externally observed responses at every protocol cut.
• FS2 — Record validity: distinguish incomplete tail from malformed complete record; explain sequence, marker and checksum limitations.
• FS3 — Implementation: own starter passes supplied checks and an independent overwrite case; explain why a retained full log permits redo from stale checkpoints.
• FS4 — Failure discrimination: exhibit lost early success and non-idempotent replay, then show repair without hiding the failure.
• FS5 — Scope and uncertainty: explain unacknowledged durable commits, process-crash versus power-loss tests, flush errors and directory persistence; do not infer actual device behavior from the model.
• FS6 — Delayed transfer: after 48 hours, solve a fresh crash cut and identify the extra protocol required for log reclamation, concurrent writers or retry deduplication.

Allocate 6–8 hours including repair and recall; elapsed waiting is not study time. Record initial score, evidence path, repair, fresh case and final score. Failed evidence means incomplete. Automated checks do not establish independent understanding. Use calibration after your attempt; existing module requirements remain required.