Only new NS-72–74 evidence is included. RH and G2 remain open. No independent review is claimed. The complete Gram/tail kernel is inherited unchanged from v159/ns61, and projection helpers from v160/ns71; hashes are recorded.
ns72/: causal factor, whole-line differential energy, interpolation and
size-independent model-curvature comparison; complete finite gains.ns73/: prescribed physical Mobius/log-taper directions and their complete
optimized three-dimensional span. This is not an asymptotic exclusion.ns74/: unitary altered family with unchanged complete Grams and an explicit
positive floor. This is not a zero or floor for the original zeta family.proof.tex: proof fragments concatenated verbatim in the live manuscript.validation.json: actual build, replay, source-hash and review scope.All finite numerical evidence is ball arithmetic; no midpoint solve is used. The finite kernel cutoff 128 retains the full Bernoulli/Hurwitz remainder, with order 24. Cutoff 256 tightens the same complete problem at N=256. All five sizes N=16,32,64,128,256 run at 256 and 384 bits. Every run has a source hash and inherited-dependency hashes. NS74 offsets are exactly 1/10^6 and 1/10^10 above 1/2. The altered problem has the same target norm 2.
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python evidence/v161/ns72/verify_replays.py
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python evidence/v161/verify_replays.py
pyright evidence/v161
./manuscript/build.sh
python3 tools/verify_manifest.py v161
Fresh calculations: invoke each certify_*.py --help, using the sizes and
precisions above. Use --cutoff 256 --sizes 256 for the cutoff replay.
Run all three exact-checks.mac files with Maxima and require saved PASS
counts 41, 10 and 16, not just a successful process status.
The local energy proof uses the additive positive derivative functional on each piece and identifies it with the L0 norm only after joining the whole-line extension. No finite-piece integration boundary is dropped. The O(N) rule counts scalar operations after the actual correlations are known; neither finding those correlations nor bit complexity is claimed linear. The minimum-accuracy summaries for NS73 additionally check all serialized fit coefficients, beyond the gain/energy metrics recorded by its generator.
No finite table supplies the cofinal curvature/numerator or true-cost bound.