Chapter 1
Collapse SimulationRun fixed-seed collapse scenarios with tick-level traces and deterministic replay checksums.
Question and Decision Rule
Restart Verdict = YES only if every hard gate passes; otherwise verdict = NO.
The leaderboard rank is not enough. A model must satisfy collapse recovery, decision integrity, fallback reliability, pass-rate, and evidence-depth gates simultaneously.
Distinct collapse models >= 4
Runs per model >= 24
Top pass rate >= 0.70
Exam Story Pipeline
Chapter 2
Hard Gate ScoringScore regen, collapse recovery, decision integrity, fallback rate, and pass-rate against strict thresholds.
Chapter 3
Reliability ChecksRun pairwise swap checks, disagreement instability diagnostics, and anti-template filters.
Chapter 4
Public VerdictPublish leaderboard plus `/api/restart-verdict` so anyone can inspect failed gates behind a NO.
Failure Conditions and Limits
- A single failed hard gate forces a NO verdict.
- Fallback-only operation is treated as a critical blocker.
- The exam measures collapse recovery behavior, not full civilizational omniscience.
Strict by design: hard to pass
Bounded to published scenario packs