S. Project Evaluation Checklist

For every project you build — Part XVIII's and beyond (Ch. 46.8 + 54 + 62's standards in one card). Correctness: does a one-command run reproduce the headline result? property tests green (unitaries, oracles, ancilla-cleanliness)? differential tests vs. an independent implementation? Honesty: seeds fixed or swept — reported either way; error bars on every claimed number; baselines at equal effort; negative results included, not hidden. Engineering: pinned dependencies; tests in CI; README states claim, cost, complexity, and how to verify; code review-clean (another engineer could take it over). Benchmarks (if performance claims): standard corpus; ≥20 seeds; medians + IQR; ablations isolating each contribution; comparison to the strongest available alternative, not the weakest. Science (if research claims): hypothesis pre-registered; regime named; noise model justified with falsifier; statistics per Ch. 54.8. Portfolio (public-facing): figures self-contained (regime, shots, error bars in caption); write-up tells a stranger why it matters in three sentences; license present; the 6-month test — will this still run? The final question: if this project's headline number were wrong, would I know? If not, that's the next task.