Independent verification: uniform U8 supporting inequality (no new kappa)

[woklerx]

PR942 verifies the U8 rank-one supporting inequality at one admitted endpoint. This independent research package extends it to every real 0 < b <= 1/3 and derives a quantitative penalty for residual ranks outside {0,1,48}. It changes no supplier, selected frontier, or kappa.

The written monotonicity argument reduces the continuum to b=1/3. Two standalone standard-library implementations check all 1,163 legal bin patterns and 42 strict inequalities, using different enumeration methods and integer cube-root formulations. Both certify a uniform scaled margin greater than 554. The resulting fractional lower envelope holds for every positive replica count; equality in the integer relaxation requires 37 | T. Owner feasibility remains separate.

Reproduce from the repository root:

python -B research/u8-supporting-uniform/test_verify.py

Expected: 8 tests, OK. Local Windows/CPython 3.14.5 validation includes both implementations under -O -S -B, nine certificate corruption controls with required rejection reasons, source-drift rejection, exact source-model binding, and byte-identical certificate regeneration. A fresh portable source-closure ZIP also passes all eight tests. The exported patch applies to the exact base using an isolated index, and the selected-record integrity check passes. Repository-wide make verify, native supplier cold replay, hosted CI, and proof-assistant verification were not run.

Base: 407d52ff15b2217b6e22bce44728b4c3907b6515. Track/seed: independent verification, OS-random seed 77. Source SHA-256 pins, exact dependency heads, completed coverage, proof, validation limits, and local claim are in research/u8-supporting-uniform/. Final overlap searches found predecessor #942 but no matching uniform-U8 package; the five pinned inputs remain unchanged at refreshed main 8bddbee377e2f3ce1f1247196c4b290d55767d20. No pinned coordination issue was visible, so the claim is local.

Credit Chafik Boukhalfa's #795/#890 prices and ledger and Jonathan Capalbo's #942 pointwise all-T bound. Charles Bethin's #850 proves a stronger integer optimum for the fixed T=8 instance; this fractional all-T result does not supersede it. Joel Pulikkan's concurrent #962 independently reconstructs T37 owners and was inspected for overlap; its code is not consumed here. Other U8 lineage credits are preserved in NOTICE.md.

This is a written mathematical argument with exact finite certificates, not a proof-assistant multiplication theorem. All inherited compiler, realization, hosting, precision/tape, analytic, and bit-interface obligations remain. The same supporting line actually fails at b=1, with an exact rank-32 counterexample included. Prepared for @woklerx with substantial OpenAI Codex assistance, under Apache-2.0; no independent human peer review is claimed.