Integration: admit PR883 X4 primitive lowering through selected invoice and assembly (no new kappa)

[nightangelflowerwin-ops]

Follow-up to Philipp (@mightyOil)'s #883: emit its four-primitive X4 precondition blocks as complete concrete programs, bind the compiler refinement to fresh selected supplier admission, and reissue the entire 72-call invoice and unchanged assembly. No new kappa.

The admitted selected bill falls by 56,448 real-component primitives (784 per call/orientation) to 13,852,752. Coefficient-work operations fall by 36,288 under the predecessor's per-component convention; they are already included in the primitive bill. Physical stock remains 7682 and prefix remains 257/512 bits. All finite guards, exact moments, bootstrap and 47 outer inequalities pass at the unchanged selected kappa 1661634653669567/1000000000000000000.

The original macro ledger is retained as the logical/native interface. The new explicit primitive stream has identical endpoints at every frame-zero pair boundary and preserves every physical phase/frame/center/wrapper event. A separate two-row matrix checker validates the actual emitted blocks; complete emitted scalar replay covers 728 native plus nine arbitrary private columns. Private coefficient work is not assumed clean. PROOF.md distinguishes this certified compiler refinement from a claim that the historical verifier parses the new primitive format.

The original source package was freshly reproduced, including its independent pair proof and seven source/interface tests. A new PR799 native/global replay passes all endpoint controls, 2949 phase edges, wrappers, owner/invoice/assembly and independent checks. It explicitly reuses complete verified PR741 ancestry, not a fresh bit or SW4 reconstruction. Vendored unchanged #865 coverage is mandatory before refinement. FRESH-REPLAY.json retains the full binding and stage evidence.

Reproduce with a complete PR799 worker output and its actual verified ancestry:

python -I -B research/x4-primitive-admission/admit.py --replay /complete/799/replay --ancestor /complete/741 --out /fresh/external/admission
python -I -B research/x4-primitive-admission/test_controls.py /fresh/external/admission

Six prefix corruptions per orientation, source drift and three assembly corruptions reject (16 new controls); the 672 predecessor pair mutants are also rechecked. Undercharge, failed precision and missing lowering contract publish no assembly result. Relocation to another complete source clone with spaces in the output path passes. Selected-record and staged source-byte/whitespace checks pass.

The broader repository suite was previously attempted and stopped on eleven ten-second subprocess timeouts in an unchanged historical test under Ubuntu WSL. No repository-wide or hosted-CI pass or runtime speedup is claimed for this refinement.

Original PR883 source/credit is preserved byte-for-byte. Credit #865's evidence gate, Lucas (@ljste)'s #838, concurrent #847, and the Jeremy Kothe/Chafik Boukhalfa/icekylinx/DreamingOfClouds/Rohan Arun/eumemic native/compiler/owner/bit lineage. Concurrent ToikY's #901 changes the subsequent compensation and preserves these preconditions; its code and savings are not consumed or added. No factorization or integration priority claim. Base is f567884aeb8851b6e794af1184119a3bfbe1e631; selected/default/upstream sources remain unchanged.

Prepared for nightangelflowerwin-ops with substantial OpenAI Codex assistance, Apache-2.0. Finite compiler/operator/accounting checks retain all-size compilation, realization/hosting, precision/tape and analytic assumptions. No full multiplication theorem or independent human review is claimed.