CORRECTED CONSOLIDATED ODD-PRIME MIXED-STATE LEARNER
Scott Schubr -- October 18, 2026 -- correction revision 1

Mixed-state escape bounds and proper agnostic learning of stabilizer states in
odd-prime dimensions

READ FIRST
manuscript/main.pdf: final nine-page paper; main.tex: complete analytic proof.
REVISION_NOTES.txt: four requested minor improvements and editorial changes.
review/REVISION_DISPOSITION.txt: current readiness and source-bound review scope.
PROVENANCE.json: exact current identities, authorization and preserved baselines.
PRESERVATION_RECEIPT.json: complete two-original preservation and ZIP readback.
MANIFEST.sha256.json: exact package identities, excluding itself; not a proof.

This is ONE consolidated paper. The qutrit theorem is exactly its p=3 special
case. Both earlier sealed packages and their ZIPs remain unchanged on D:.
Baseline/ subdirectories contain historical evidence for earlier source editions;
their old hashes, permissions and page counts are historical, not current claims.

MATHEMATICS
Every odd prime p, n>=1, arbitrary mixed rho. For any globally maximizing
stabilizer label Lagrangian S and proper LINEAR T<S, the physical escape law
has mass >=3(p-1)F^6/(4p^3) using six copies for p=3 mod4, and
>=(p-1)F^4/(2p^2) using four copies for p=1 mod4. These are sufficient counts.
The proper learner has output overlap >=F-epsilon with confidence >=1-delta.
For L=ln(2/tau), its conservative resource bound is
exp(O(p^2 L(L+ln p))) poly(n,p,1/epsilon,ln(2/delta)); fixed p is
quasipolynomial in inverse overlap. Exact Clifford and coherent measurements
are the operation model. tau=epsilon removes the overlap promise.
No theorem, constant, algorithm or resource formula changed in this revision.

CONTRIBUTION AND LIMITS
The proposed increment is the mixed purity-sensitive local estimate, its
all-prime operational escape consequence, and the complete mixed-input learner
analysis. MUB identities, Fourier-dual kernels, weighted/skewed measurements,
arithmetic split, stabilizer algebra and bootstrapping are credited to prior work.
Current primary-literature comparison supports provisional specialist novelty
and value; it does not certify worldwide priority. No optimal-resource,
practical-efficiency, composite-dimension or finite-universal-gate robustness
theorem is claimed. AI-assisted internal review is not external human refereeing.

FRESH REPLAY
verification/REPLAY_RECEIPT.json binds revised executable sources to fresh runs.
Full suite:178 density inputs,36 physical batch laws,1010 exact Fraction grid
identities. Dense independent p3/p5 PVMs and a distinct weighted p7 PVM pass.
Ordinary Python, -O and inherited PYTHONOPTIMIZE runs pass; deliberately false
escape/dense gates reject. Manifest corruption/unsafe path controls reject.
Finite complex128 diagnostics do not prove arbitrary n or simulate the learner.

REPRODUCE (Python3 + NumPy; integrity alone uses standard library)
 python -B verification/verify_package.py
 python -B verification/verify_package.py --dense
 python -B verification/verify_package.py --full
 python -O -B verification/verify_package.py --full
 python -B verification/verifier_robustness_checks.py
 python -B verification/independent_p7.py
All wrappers above place fresh outputs in new external temporary directories.
The p7 dense diagnostic uses about100MB for its projector arrays plus overhead.
Full-suite enumeration is implemented only for the documented n=1 and n=2 cases.
Do not run the two low-level output-producing diagnostics inside a sealed folder;
use the wrappers or copy those scripts together to separate scratch first.
Compile manuscript/main.tex with Tectonic or standard LaTeX, writing build outputs
to separate scratch when preserving the sealed package.

AUTHORIZATION
Scott explicitly authorized this corrected package's delivery to Luminary Research
Archive for publication. Publish one consolidated work; do not release the qutrit
predecessor as a separate paper. Handoff, observed publication and verified public
download acceptance are separate states. See the external delivery receipt for
the actual release outcome; no receipt is inferred merely from these files.
