[aliromman]

The actual 112-port Gaussian reader reaches character rank 30, so its sharp dirty-bank cost is 2 ceil(30/q)+2 cross-bank shears. An explicit dyadic centre basis reduces the four-shear word from 31,608 to 20,352 coefficient ADDs (35.6%), without a new odd denominator. This closes the actual-source rank and coefficient-cost gap left by PR5640 and PR5634. No new kappa is claimed.

The proof and boundary, basis witness, all-width storage tradeoff, exact verifier, source manifest and frozen context are pinned to contribution head 98c17101e47ad378702f8c3eb450a87b7ac7bdc2. Canonical base and final refresh are df0137efd35e11b4445734bf612bdd4c7ebc330c, priority revision 7, selected conditional kappa 0.00310045. Only this package is added. Coordination.

All 512 source character inverses, reader identities and exact rational ranks pass; modular ranks give matching lower bounds. Character 92 has rank 30. The physical inverse Walsh kernel has 15,628 reader coefficients. Ten basis changes reduce this to 9,968, with 208 scatter coefficients. The common denominator is 960, with odd part 15. The complete search over all 870 ordered row pairs and every possible improving dyadic coefficient proves that no single further row ADD reduces the bill. This is a local basis optimum, not a global ordinary-cost optimum.

The package emits exact coefficient bills and coefficient-ADD prefix/suffix bounds for every borrowed dimension 1 through 30. One borrowed field uses 62 shears and 18,331 coefficient ADDs; thirty use four and 20,352. All translation shifts and rational coefficients remain charged. A separate literal 512-address dirty-input replay agrees with the original source factors. Three controls reject a corrupted actual inverse, a wrong restoration-gather sign and a missing final scatter. The all-input endpoint follows from the checked factor identities and the written word proof; the physical test uses one deterministic payload.

python3.12 -B scripts/research_package.py --package research/actual-gaussian-reader-tariff
python3.12 -B scripts/check_research_commit.py --package research/actual-gaussian-reader-tariff

Expected: PASS_ACTUAL_GAUSSIAN_READER_TARIFF, PASS_REGISTERED_RESEARCH_PACKAGE, and PASS_RESEARCH_COMMIT_BYTES. Normal and optimized registered checks, exact committed-byte checks, export validation and a Git-free optimized replay pass. Python 3.12 and the standard library suffice. CI is pending at publication. Full-history preflight is unavailable in this shallow checkout; source hashes and committed bytes were checked.

The lower bound uses the translation-kernel cross-bank alphabet. Unrestricted address mixing has a different contract. A smaller borrowed bank does not remove the thirty-field bank used inside the section. Rational arithmetic interiors, physical routing/tape costs, native/outer admission and a complete supplier/caller remain open. No full historical suite, selected cold replay, proof assistant or independent human scientific review was run.

Credit David Ganz (@Dugongue) for PR5634 and its native/campaign lineage, PR5640 for the sharp tariff, and centre-code-services for the actual response and inverse certificate. Frozen notices retain reciprocal-native-responses, native-centre-capture, centre-metric-encoders, PR3133, PR3148, PR2588 and PR2226. New source binding, basis reduction, proof and checks: Ali Romman with substantial OpenAI Codex assistance. Apache-2.0. Next questions are a broader physically priced basis and one complete native caller; this package reserves neither.