[thu-nmrc]

Result

The frozen native full-frame histories have an explicit input kernel at every positive tensor width. In logical native input coordinates, all backward observations are unchanged by (X,Y,H) -> (X+v,Y+v,H); the forward observations are unchanged by Y -> Y+v. All helpers and owned inputs stay fixed. Thus the complete full-frame channel cannot exactly determine any nonconstant function of arbitrary original Y, even with arbitrary nonlinear postprocessing or unrestricted address-operator left combinations.

This strengthens the earlier scalar/finite-image result in this same PR. It is a source-specific information theorem with no new κ, paid profile, profitable caller or selected-frontier change.

Frozen native body	Distinct nonzero full-frame histories	Kernel
#5402 raw backward	4,891	[I; I; 0]
final #4610 raw backward	5,019	[I; I; 0]
final #4610 raw forward	4,968	[0; I; 0]

There are 112 X, 112 Y and 408 independent helper fields. The small backward certificate has 224 unit entries. The decisive identities are H_f K=0 and K_Y=I; the backward channel may depend on X−Y, so this does not assert independence from Y when X is fixed.

Review object and proof

The exact publication head is be292a4c2724a17c7401bb736f969e898c43f0f2. Review KERNEL-PROOF.md, KERNEL.json, KERNEL-RESULT.json and the separate implementation's audit.

• Fresh actual signed-native replay checks cancellation coefficient by coefficient in all three complete histories. No monomial-matrix independence, modular nonmembership, width-one sum relation or operator-polynomial inverse is needed.
• A separate native/group/frame implementation derives the provider kernel using exact Fraction elimination of scalar augmentations, then decisively verifies it against every original operator-valued history row. It imports neither the primary interpreter nor monomial/discovery code. Saved kernel rows are compared only after derivation.
• An alternate formal-word algebra reproduces all three identities using 72 free generator names and only adjacent inverse cancellation. It shares the source event interpreter and is explicitly corroboration, not an independent parser. Scalar coefficients remain outside tensor powers.
• Physical input displacements use the corresponding invertible birth frames. The common raw-backward full-frame factor is R inverse, with the retained conjugated letters. A zero logical history difference therefore remains zero physically; the argument does not claim free physical input preparation.

The nonidentifiability argument is ordinary linear algebra: input instances z and z+Kv have identical observations while Y varies arbitrarily. The contribution is the source-bound complete-channel identities and physical interpretation. The 13 MB module decomposition used during discovery is not shipped and is not a proof/runtime dependency.

Exact scope

The observation data are only the retained full-nine-dimensional-frame values of each pinned native body, considered separately, including all versions after an intended exit. Original X/Y, partial-frame values/caches, extra paid transformations and changed words or input relations supply different information. Fixing helpers or owned values alone does not remove this kernel.

In particular, raw #5402 WORD ends with eight nonzero partial-frame output caches, before the provider boundary exchanges/clears them. They are excluded from this channel; the whole machine state is not claimed to hide Y. A provider-boundary extension must check any additional values explicitly. #5413's neutral completion is a changed paid-history resource and is not excluded here. The target b=ZQY_return uses the specified native return Y_return=Dy; a later caller Y contaminated by X is a different target. No finite-precision approximation lower bound is proved.

The earlier scalar/Gram/finite-ABI scripts and results remain byte-identical to initial head 6d81d47e6d107e3c52b4583526d31e102b219bdc and remain optional direct replays. Their restricted numerical statements are not dependencies of the new kernel proof.

Reproduction and validation

Python 3.12, standard library only:

python3 -B scripts/research_package.py --package research/physical-zqy-scalar-span

All eight registered commands pass, normally and with optimization: direct kernel, separate implementation, formal free-word corroboration and retained exact Clifford phase audit. The phase audit compares 147,456 native columns and 36,864 product columns. Explicit failure controls remain active under -O.

Committed-byte preflight passes for 31 package/input files at this exact head. Official export and archive checks pass. All eight commands also pass in the exported read-only, Git-free directory (32 files including the export receipt), with files mode 0444, directories mode 0555, no Git on PATH, no .git directory, and identical pre/post inventory and SHA256 hashes. Archive SHA256: 6353e05f53e07c09bcb9df4996528305a8559a429b067455f7091a9b6b0ecbb2. Local runtime is Python 3.12.13. The checkout is shallow; optional full-history ancestry preflight was not run. Direct base/publication ancestry passes.

The hosted registered-package workflow passes at this exact final head. The final follow-up commit only corrects the recorded local Python patch version; its scientific scripts and certificates are identical to the preceding proof commit, which also passed CI. This is reproducible replay evidence, not external scientific acceptance.

No selected cold replay, fresh complete native admission, compiler/outer, precision/tape/stock, Lean, Docker or complete multiplication-theorem audit is claimed. Internal AI-assisted corroboration and replay success remain distinct from external scientific acceptance.

Sources and credit

Publication base is canonical main 32890bc5de034ab8fc93ac88ead7279a6a416d04, refreshed before submission. It selects conditional κ=.003027288235 from #5413/#5414. Five existing canonical runtime inputs are pinned by Git blob and SHA256 in SOURCES.json; frozen inputs and selected-frontier files are unchanged.

Prepared by THU New Media Research Center (@thu-nmrc), with substantial OpenAI Codex multi-agent assistance; Apache-2.0. Preserve Mark/Matarener's #5427 question and event/version semantics, Dillon Amburgey's #5402 provider, Cao Yifan's #4610 native word, zxv's #5415/#5419 observer/consumer ABI, and Cinna Davis, Dugongue, Ali Romman, Jacob Alan Sussman, Chris R/Clear1nsight and all retained native/phase/controller/compiler notices linked in README.md.

The next constructive task must add genuinely different information: an exact partial-frame or changed paid interface with current-version ownership, all independent original inputs and a complete favorable observer/termination bill. Further postprocessing searches on this same frozen full-frame-only channel are stopped.