Incremental NS-75–77 evidence only. RH/G2 remain open. Local self-review only. The classical factorization and published zero verification are external inputs.
ns75/: exact divisor-log cells and full jump measure; actual coefficient
feedback, three-direction span, and held-old unit-step control. Every finite
quantity uses the full arithmetic Gram, tails, projection and cross terms.ns76/: complete tail unitary, exact continuous q=2 distance, invisible
residual component, and strict closure separation with a certified witness.ns77/: positive complete zero-tail bound and scalar calibration. The
Platt–Trudgian zero verification is cited, not re-executed or archived here.proof.tex is integrated verbatim in the live manuscript.PYTHONDONTWRITEBYTECODE=1 .venv/bin/python evidence/v162/verify_replays.py
pyright evidence/v162
./manuscript/build.sh
python3 tools/verify_manifest.py v162
For fresh NS75 runs use certify_divisor.py --bits 256 --sizes 16 32 64 128 256
--output <new-path> and replay at 384 bits. The cutoff replay adds
--cutoff 256 --sizes 256. The default cutoff 128 and Bernoulli order 24
retain the complete Hurwitz/Bernoulli tail; increasing cutoff tightens the
same problem. Inherited kernel/projection/curvature hashes are recorded.
NS76 certify_local_gap.py --bits 256 and NS77 certify_tail.py --bits 256
print JSON; repeat at 384 bits. Their complete elementary expressions have
no numerical cutoff. NS77 handles the full zero tail analytically.
Run the three Maxima files; require PASS counts 11, 29 and 18.
No midpoint solve is used. Source hashes, replay enclosures, manuscript intervals, overlap with the published curvature calculation, and claim scope are checked. The local separator is a lower bound for sigma_(3/2), not tau. The tiny continuous bound is not an integer error upper bound.