[aliromman]

This scoped negative result closes eight proposed main-bank specializations of the retained native source. Each exact dependency slice costs more than its complete target list for every tariff H_r = r^t with 0 < t < 1, even when all needed physical repairs are free. The complete kappa calculation does not improve.

The proof, exact receipt, and source pins state the class. Base: 32890bc5de034ab8fc93ac88ead7279a6a416d04. The package changes no predecessor word or selected record.

Set one main bank, or both, to owned zero. Retain all 408 independent helper outputs and the specified main output bank. The checker independently replays the source rows, removes proved-zero paid calls and events outside the backward dependency slice, and replays the retained sequence. It also checks two forward variants that omit the 6,037-record prefix before entry_adapters. Every helper target remains credited.

The eight target-minus-paid rank surpluses are 0,-842,-195,-131,0,-848,-881,-704. Rank surplus alone is not the proof. Each case has nine nonnegative signed hinge values and a positive first hinge. The zero-surplus X-to-X cases have the exact certificate [2283,2615,2680,2275,1795,1303,811,326,0], which still gives a strict deficit for every stated power. Both prefix variants have the same paid profiles as their corresponding full-prefix slices.

The result covers these deterministic slices only. It does not prove optimal pruning, forbid changed scalar words or physical identities, or admit a sliced native supplier. A different caller or output contract needs a new invoice. The next experiment is to supply such a changed word and verify its complete cost.

Cao Yifan supplies PR #4610, its frozen word, and its source attribution, which preserves Cinna Davis, Dugongue, zxv, Jacob Alan Sussman, Pierre-Loic Doulcet, and the retained lineage. PR #5397 supplies the shared signed-invoice checker. Its classical concave-order identity is reused, not claimed as new. This package changes the main-bank input domain; it does not repeat that PR's eight helper-relation census. New analysis, proof, and checks: Ali Romman, with substantial OpenAI Codex assistance. Apache-2.0.

Use Python 3.12 and the standard library:

python3 -B scripts/research_package.py --package research/native-main-bank-slice-gate
python3 -B scripts/check_research_commit.py --package research/native-main-bank-slice-gate

The direct verifier prints:

PASS_NATIVE_MAIN_BANK_SLICE_GATE_NO_NEW_KAPPA

Registered normal and optimized checks, committed-byte and export-integrity checks, and normal/optimized replay from a fresh Git-free export passed. Exact controls reject changed source bytes, a wrong sign in the last helper ADD, a missing independent helper output, and unearned target credit. A positive concavity case checks the comparison direction. A separate Codex worker reviewed the proof, reconstructed all ten hinge vectors and all helper credits, and replayed the full and sliced X-to-X case; no blocker was found. This is AI review, not independent human acceptance.

The optional full-history check was not run in this shallow checkout; the publication commit directly follows the declared base. Physical payload, absolute phase, frame repair, compiler, precision/tape, and outer multiplication checks were not run. Hosted CI has not run at publication.