[tekuuu]

Change and mathematical scope

Construction / implementation simplification. PR3131 turns each reversible dense address word into a protected XOR oracle, then uses two such oracles to make an in-place gate. The word already preserves a known scaled Gaussian integer lattice. Executing it directly and contracting its zero guards cuts the actual paired collector's wrapper-level lifting calls from 416 to 104, with four guard embeddings/contractions instead of 104. Both original nonlinear cores and all internal backend cleanup remain paid.

The complete query, adaptive grid and private-zero return agree exactly with the pinned predecessor. The same-input N=512 Python semantic collector measures 2.62–2.66x faster across four fixtures, with three samples after warmup; all samples and tested source hashes are included. This does not execute the fast physical C trace or change a supplier rank histogram, asymptotic exponent or κ. The physical/compiler/precision premises and caller clean-stock hosting remain open; the frozen consumer's 599/584 zero-saving obstruction is unchanged.

Task and overlap

Base: 997e6a9ca7b4f95d50e366f61412fd640d72a88a, priority revision 6. Dependency PR3131: 0c9e344d5a5295a218558fa4a51d867f39ff5311; its frozen PR3109 backend: 15eac811648182bdac7254a91fd9b841b1414ebc. The package authenticates the unchanged eleven-file archive and declares all eleven external replay inputs.

OS seed 78 assigned independent verification. The independent-moment card scope is explicitly changed to a constructive, independently checked wrapper-boundary refinement. The new code checks the affine compiler output symbolically and provides separate parity/address oracles; the inherited backend and collector are reused, not claimed independently validated. PR3122/3129's history work is complementary and is not composed here. Final refresh covered 258 open PR metadata records, current predecessor feedback and the collector discussion; new PR3133 concerns a different carrier-fold construction.

Canonical claim: #2750 (comment)

Validation

python3.12 -B scripts/research_package.py --package research/guarded-wrapper-refinement

Expected: PASS_GUARDED_WRAPPER_REFINEMENT wrapper_lifts=416->104 collector_cases=4 and PASS_REGISTERED_RESEARCH_PACKAGE outcome=completed.

Registered Python 3.12.14 replay passes. An optimized, Git-free Python 3.14.7 replay reproduces the deterministic receipt exactly. Checks include exact Q(i) identities, 1,250 signed pair cases, 13,122 exhaustive gate cases, 9,144 signed bit embeddings/contractions, 196 phase-changing bit outputs, sixteen actual-wrapper and sixteen tensor cases, and four complete collectors. The tensor cases include a genuinely changed inner stage order. Controls cover dirty guards, wrong sign extension, overflow, unguarded wrong endpoints, phase/inverse/word/layout/type faults and source corruption. Publication-byte preflight passes.

The parameterized argument is in PROOF.md; finite replay is not a formal proof of the full theorem. No full supplier/outer cold replay is claimed or required for this isolated wrapper refinement. Hosted CI is separate from these local checks.

Credit

Tekalign, with substantial OpenAI Codex assistance. Apache-2.0. Credit Jose Manuel (jmbarrancoml)/PR3131 and the PR3109/2287/2250/2899 implementation chain; PR3025/3033; Ross Richardson/PR3072; Douglas Colkitt's character conjugation; ywz65's native recipes; Vasek/flyingviper's consumer audit; Nick Petalas, David Arias and the retained source lineage. Original credits and source bytes are preserved. No new general reversible-computation theorem, independent human review or proof-kernel verification is claimed.

Open next questions

Lower one direct wrapper to the actual physical backend with literal guard/sign backing and private-history return, then price an enclosing caller's complete clean-stock and consumer contract. README.md and TASK.json give the first steps and blockers. Fewer wrapper calls alone do not repair the frozen supplier invoice.