[GoGoKo699]

Scope

Implements the two software issues found in the 7 September takeover review. The August scientific correction, PDF/TeX, witness proofs, result files, AI-assistance acknowledgment, six historical Python scripts, and immutable historical release are unchanged.

Changes

• Stable scaled-recurrence binomial evaluation and natural-log tail APIs, without a new runtime dependency.
• Positive probability underflow raises explicitly rather than silently returning zero.
• Strict validation of all five result records and all 458 fields, using expected values built without reading the files under test.
• Deliberate corruption of each JSON leaf and CSV cell, including duplicate-record tampering, missing fields, NaN/infinity, malformed schemas, and Python -O checks.
• Complete SHA-256 manifest verification, including source ZIPs.
• Python 3.10, 3.11, 3.12, and 3.13 CI matrix. Version 1.1.1.

Validation completed locally

• 59 tests passed, including 458 individual result-field mutations.
• 65 file checksums and all six historical Git blob hashes passed.
• All 458 scientific-result fields validated.
• Independent rational references cover the former overflow cases: C=1500,E=300 and C=2000,E=400.
• An additional 239 finite-log-CDF comparisons against SciPy 1.17.0 agreed; maximum observed absolute log discrepancy was 5.85e-12. SciPy is not a project dependency.
• The uploaded Git tree is exactly the locally tested tree: 8196bcc3e88ce21a9d6b057a5208f2a14e33b7da.

Explicit limitations

Fast validation checks archived Monte Carlo counts, metadata, and recomputed Wilson intervals; it does not rerun the original million-trial experiment. Floating-point probabilities are evaluations of the analytical bound, not interval-arithmetic certificates. No scientific claim or experimental trust assumption is broadened.