[Gonzalo-Mier]

Change and mathematical scope

Constructs one explicit native P112 co-delivery block rather than pricing its coefficient supports alone. General canonical Clifford frames realize four independent query transfers at rank one, with one explicitly cleared owned scratch array. The source movements cost 3+5; framed port delivery also costs 3+5. Exact native recipes and complete local histograms are retained.

Four separately specified output contracts pass on arbitrary data. A frame-zero query return has rank four; it cannot inherit the framed rank-one price. The package also distinguishes the abstract transposed reader from the actual destination-first selected scatter, and executes an actual-scatter/product control with unchanged centre inputs.

This is a constructive local module and a scoped boundary-cost result. It does not repair the changed port endpoint, supply a selected chronological producer/consumer, prove a second unpaid coordinate, or improve κ. Selected κ remains 0.001672660197076918.

Task and overlap

Base: freshly fetched canonical main 20ecf2891d80b5acf18d980b2164fb44ea7faa90. Task complex-rank-shape-decoder, changed scope to constructing the first co-delivery block under the donor's request for mechanisms crossing κ=0.002.

Builds on PR3566 and PR3604. PR3691, head b37d0e75433837482c351863a672543056f95bd5, already checks abstract composition and a losing port repair, while explicitly leaving native rank-one COPY unconstructed. This PR adds that local constructor; it does not refute the port defect. PR3632, head 7e7a889cbf6da2c4461af806fe52a888c621a7e7, identifies the scatter orientation, which the actual-scatter control preserves.

The overlap refresh covered 484 open PRs and 2508 Discussion records with complete pagination. Both named audit heads were rechecked before submission. This is not a novelty certificate. Exact source pins and the original card are in the package metadata; replay requires no pending-PR source files.

Validation

On CPython 3.12.15 in an isolated checkout:

python3.12 -B scripts/research_package.py --package research/p112-codelivery-native-block

Normal and optimized registered replays passed: 25 independent scalar columns in each of four module contracts, 23 exact native recipes with separate Gaussian phase/Pauli checks, every one of the rank-one copy's 512 basis columns, scratch lifetimes and local cost ledgers. Controls reject wrong centre signs, reversed scatter operands and wrong global phase. The fixed-zero rank-four operator is retained as a boundary control.

No full selected-supplier cold replay, physical hosting/precision admission, complete bit improvement or Lean proof is claimed. The tensor extension of the local operator identities is described separately from those missing obligations.

Credit

Gonzalo Mier's P112 coefficient/consumer work and the credited selected geometry, word and compiler lineage; Ross Richardson's scatter correction; the selective co-delivery audit; and the retained P104 exact Clifford compiler and separate Gaussian interpreter. Prepared by Gonzalo Mier with substantial OpenAI Codex assistance. Apache-2.0; no independent human review claimed.

Open next questions

Bind one actual selected producer and suffix to a specified module contract, replay the complete changed port/helper endpoint, and price source acquisition, the owned scratch array, address/layout work and precision. The existing rank-one recipe does not discharge that boundary or the independent bit-side κ requirement.