[CubeJerry]

Change and mathematical scope

Proves deterministic quantizer transparency for the complete selected #6030 caller, for every integer f>=1 and arbitrary correlated finite spectators. If input real/imaginary components lie on the p-grid with magnitude below 2^h, the sufficient format is F=p+ceil(145f/2)+4 fractional bits and W=p+h+145f+20 signed bits PER component; W+6 covers numerator-first arithmetic.

The source certificate covers scalar products, all 18 local calls, 16 reverse interfaces, helpers, queries and completed direct C stages. Local coefficient mass is at most 850 (504 before addition); all 363 global events have integral prefixes of mass at most 10. Completed common-chart endpoints reset the four-bit exterior debt.

No new κ. This does not set an accelerated native child's internal width P_r to W, prove its return exact, or implement physical owned erasure.

Sources and reproduction

Base: 7c49fe32c55de02c2c873655c804b16e4c3fef0a. SOURCES.json pins 13 canonical input/provenance files. DEPENDENCY.json pins four byte-identical operator artifacts from published e56a5bcf239fa0cd4ee57459af04c0ba7acb2c03.

Credits #2725's exact-grid lineage and #2287's existing recursive common-grid method. #6499 treats the current local word with independent native-return atoms; this checks actual correlated charts and the complete caller. #6635's scalar lift remains credited. No new Clifford denominator lemma is claimed.

python3 -I -Werror -B scripts/research_package.py --package research/selected-caller-quantizer-transparency

Expected: two PASS_SELECTED_CALLER_QUANTIZER_TRANSPARENCY markers and PASS_REGISTERED_RESEARCH_PACKAGE.

Validation and remaining work

Executed normal/optimized registered checks, committed-byte/source verification and independent valuation, backward-Laurent, canonicalizer and Fraction-kernel reviews. The complete 22,803-event local fixture checks 6,094,848 C component divisions, 19,750,912 scalar products and 394,240 public complex outputs. All 29 adverse controls reject, including lost phases, spectator contamination, wrong grids, termwise C division and signed overflow.

The written proof supplies the all-size theorem. The 512-coordinate fixture is not a global 2^145 physical replay. Thirty nonzero ideal CLEARs retain their disposal obligation. Actual P_r, compiler-wide precision, routing, entry/return conversion, warm backing and the complete paid tariff remain separate. A JSON key-type receipt error was corrected and both modes rerun; scientific RESULT bytes were unchanged.

Credit

Prepared for @CubeJerry with substantial OpenAI Codex assistance. Credit DreamingOfClouds, Ross Richardson, Ali Romman and the retained P104/native/controller/exact-grid contributors. Separate AI review is not human or proof-kernel acceptance. Apache-2.0.

Proof, exact pins, receipts and open questions.