This index is internal repository documentation. The three readable research reports are versioned HTML files, preserved with their original evidence:
These are audits and diagnostics, not new cofinal estimates or interval certificates. They add no theorem node or manuscript version. NS92–98 and NS100–101 retain their scoped open inputs; neither whole family is closed.
The original files are copied byte-for-byte, including the unsuccessful first Maxima harness output in the prime-scaling audit. Later successful output is explicitly identified in its validation record. Fields saying “local only”, “no push”, or “not published” describe the original test turns, before Alex authorized publication on September 29. They do not describe this publication. Historical paths are provenance, not portable execution instructions. No secret material is needed for replay.
publication-validation.json records this integration’s replay, checks and
file hashes. Original author checks are not independent mathematical review.
The reports’ HTML was parsed but browser rendering was not verified.
Use a Git clone retaining reviewed commit
f32cadfb7be523a8920b19be57d40d652e39a2ef, Python and mpmath==1.3.0.
From the repository root, with that dependency installed in .venv:
.venv/bin/python audits/nightly-2026-09-29/replay.py prime-scaling --dps 50 --output /tmp/prime-50.json
.venv/bin/python audits/nightly-2026-09-29/replay.py prime-scaling --dps 80 --output /tmp/prime-80.json
.venv/bin/python audits/nightly-2026-09-29/replay.py jacobi-product --dps 50 --output /tmp/jacobi-50.json
.venv/bin/python audits/nightly-2026-09-29/replay.py jacobi-product --dps 80 --output /tmp/jacobi-80.json
maxima --very-quiet -b audits/prime-scaling-admission-2026-09-29-v1/check_identities.mac
maxima --very-quiet -b audits/jacobi-product-admission-2026-09-29-v1/check_identities.mac
Choose fresh output paths: replay refuses to overwrite existing evidence. The wrapper changes only each original script’s Git-source location; its pinned source commit, calculations and output fields are unchanged. The cancellation audit is a theorem-applicability assessment, not a numerical experiment; its precise primary-source references and missing hypotheses are in the original JSON record.
Alex’s September 29 instruction was: “Deploy findings after every test. Allowed. Deploy tonight’s findings and then decide on the next route.” AGENTS.md now records that authorization. Completed failures are published with their limits, and the research proposal gate remains in force.
No new research route is admitted by this publication. The next-route assessment follows publication and must state its own wall check.