FINAL PREPARATION LITERATURE REFRESH AND PRECISE ATTRIBUTION
2026-10-18, separate from mathematical verification and Publisher assessment.
Only this new literature directory was written. Earlier sealed qutrit sources
and the previous research proof were preserved.

SEARCH / ACCESS
Actual Google search was freshly performed and its visible results read in
Chrome: "mixed" "odd prime" "stabilizer" learning.
https://www.google.com/search?q=%22mixed%22+%22odd+prime%22+%22stabilizer%22+learning
Relevant primary hit: Qi. Other visible hits concern stabilizer-code algebra,
Clifford rules, or resource theory, rather than an equivalent complete learner.
An initial IAB attempt failed because that browser was unavailable; Chrome
succeeded. A separate web retrieval of Google failed and was not counted as
successful Google access. Generic web searches additionally used mixed-qudit
agnostic phrasing, the Qi identifier, and purity/stabilizer-fidelity terminology.
Those tools' engine identity was unspecified and they are not labeled Google.

Primary version records and relevant full texts were reopened for the five
central papers. Backward coverage followed Qi's cited predecessors and the
qutrit baseline. The earlier actual Google Scholar exact-title query found the
Qi record without a visible cited-by link; no complete forward graph is claimed.
No equivalent complete arbitrary-mixed-input odd-prime proper learner was
identified in the inspected sources. Absence from these searches is not a
worldwide priority certificate or an exclusion of unpublished/unindexed work.

VERIFIED BIBLIOGRAPHY AND CLOSEST CLAIMS
[Qi] Wentao Qi, Boyan Xu, Shiguang Feng, Lvzhou Li.
Agnostic learning of qudit stabilizer states.
arXiv:2607.15559v1, 17 July 2026. Current abstract record still v1.
https://arxiv.org/abs/2607.15559
https://arxiv.org/html/2607.15559v1
Theorem 7: pure n-qudit input, odd-prime d, a globally optimizing stabilizer
Lagrangian S, every proper T<S, and escape >=(d-1)F^6/(4d^3).
Theorem 13 supplies the full pure-input proper learner; Section IV the bootstrap;
Section III-A the single-qudit MUB argument. Section VI explicitly identifies
mixed-input distribution and anticoncentration as extension obstacles.
The consolidated paper extends the input scope, rather than inventing qudit
agnostic learning. The mixed purity-sensitive local estimate is its relevant
analytic increment. Do not carry a pure-state local fidelity floor into mixed
conditioned blocks without proving the new inequality.

[Chen] Sitan Chen, Weiyuan Gong, Qi Ye, Zhihan Zhang.
Stabilizer bootstrapping: A recipe for efficient agnostic tomography and magic estimation.
arXiv:2408.06967v3, 5 December 2024. Current version still v3.
https://arxiv.org/abs/2408.06967
https://arxiv.org/html/2408.06967v3
Corollary 6.3 concerns arbitrary mixed n-qubit rho and outputs a pure stabilizer
with overlap at least F_S(rho)-epsilon, with high probability and quasi-polynomial
inverse-overlap resources. Lemma 5.3/Theorem 5.5 and Section 6 provide mixed escape
and bootstrapping methods. Mixed stabilizer learning itself and the architecture
are established prior work. Their approximate-local-maximizer/list results are
stronger in that separate direction; the proposed odd-prime result must not
implicitly claim those extensions. Qubit embedding and inaccessible purification
do not yield an immediate proper p-ary learner.

[Allcock] Jonathan Allcock, Joao F. Doriguello, Gabor Ivanyos, Miklos Santha.
Reconquering Bell sampling on qudits: stabilizer learning and testing, quantum pseudorandomness bounds, and more.
arXiv:2510.06848v3, 25 August 2026. Current version still v3.
https://arxiv.org/abs/2510.06848
https://arxiv.org/html/2510.06848v3
Skewed Bell sampling, register mixing, and the sums-of-squares construction are
known. The introduction and Corollary 44 already distinguish the p=3 mod4 and
p=1 mod4 cases, using a square root of -1 in the latter. The weighted commuting
Weyl construction is closely related arithmetic/measurement algebra. Credit this
precedent; do not advertise the congruence split or four-copy primitive as new.
The precise proposed mixed distribution identity and mixed escape consequence
still require the paper's proof. The HTML wording near Corollary 44 must be read
with its coordinate/scaling convention; do not copy an unqualified state-equals-
Pauli-times-conjugate assertion out of that context.

[Hinsche] Marcel Hinsche, Jens Eisert, Jose Carrasco.
The abelian state hidden subgroup problem: Learning stabilizer groups and beyond.
arXiv:2505.15770v3, 27 May 2026. Current version v3; resembles published version.
Journal record: PRX Quantum 7, 020337 (2026), DOI 10.1103/6frk-891j.
https://arxiv.org/abs/2505.15770
https://arxiv.org/html/2505.15770v3
The hidden-stabilizer-group task assumes actual symmetry and a correlation gap;
shared-code-space variants are not arbitrary mixed proper approximation. The
introduction explicitly expects qudit generalizations of agnostic learning from
its Fourier-sampling routine, giving important conceptual precedent. That
expectation is not the fully proved mixed-input optimizer escape and proper
learner claimed here. Cite if explaining the general Fourier viewpoint.

[Noller] Jan Noller, Viet T. Tran, Mariami Gachechiladze, Richard Kueng.
An infinite hierarchy of multi-copy quantum learning tasks.
arXiv:2510.08070v2, 1 September 2026. Current version still v2.
Author spelling in the source is Jan Nöller (LaTeX N\"oller).
https://arxiv.org/abs/2510.08070
https://arxiv.org/html/2510.08070v2
Section 4 measures commuting tensor-power Weyls on arbitrary density inputs,
using generalized/coarse-grained Bell measurements, to estimate Weyl amplitudes.
This is direct prior art for the qutrit three-copy primitive; two independent
outcomes followed by subtraction is standard characteristic-moment algebra.
The weighted generalization uses the same commutation mechanism. Its amplitude
learning task and copy lower bounds concern a different output requirement and
cannot automatically be transferred to the stabilizer optimization task.

STANDARD SOURCES
[MUB] Andreas Klappenecker, Martin Roetteler.
Mutually Unbiased Bases are Complex Projective 2-Designs.
arXiv:quant-ph/0502031v2, 11 February 2005.
https://arxiv.org/abs/quant-ph/0502031
https://arxiv.org/pdf/quant-ph/0502031
Theorem 3 states that a complete set of d+1 MUBs gives a complex projective
2-design. This is prior design geometry, not a new ingredient. Applying its
second-moment identity to a mixed density retains Tr(rho^2); the proposed local
escape inequality remains a further argument and must not be attributed as
though this source states it. Use the source's note acknowledging Zauner for
the earlier theorem when making historical claims; no priority claim is needed.

[Clifford] Erik Hostens, Jeroen Dehaene, Bart De Moor.
Stabilizer states and Clifford operations for systems of arbitrary dimensions, and modular arithmetic.
arXiv:quant-ph/0408190v2, 22 February 2005.
https://arxiv.org/abs/quant-ph/0408190
https://arxiv.org/pdf/quant-ph/0408190
Sections II-IV provide modular/symplectic Clifford representation and gate
decomposition. Cite for stabilizer tableau and Clifford normal-form background.
These are standard algorithmic ingredients; no new tableau formalism is claimed.

BOUNDARIES / PITFALLS FOR FINAL PACKAGE
1. Main candidate contribution: purity-sensitive mixed local inequality,
   mixed-input proper-subspace escape, and composition into a complete proper
   learner for each odd prime. Known arithmetic, Fourier/Weyl measurement,
   MUB design identity, tableau methods, and bootstrapping must remain credited.
2. The qutrit source is the p=3 predecessor, retained in the same result family.
   The consolidated paper supersedes its theorem scope without overwriting it.
3. A(x)=D^{-1}Tr(rho W_x rho W_x^dagger) sums to ONE by Weyl twirling.
   p_rho(x)=D^{-1}|Tr(rho W_x)|^2 sums to Tr(rho^2). A prior review confused
   these; that review was corrected transparently. Do not repeat that error.
4. Label exact global optimization, arbitrary density input, odd prime, n>=1,
   ordinary independent copies, and additive pure-stabilizer overlap explicitly.
5. Distinguish fixed-prime resource claims from uniform dependence on p;
   primitive copy count from total sample complexity; postselection success
   probability from the number of conditioned copies demanded by a subroutine.
6. This literature review does not certify the final inequality constants,
   analytic proof, implementation, verification scripts, PDF, or package binding.
   Fresh Publisher verification and Scott's release permission are separate.

ASSESSMENT
Relative to the versions above, a proved arbitrary-mixed odd-prime escape and
proper learner is a plausible substantive scope extension, with significant
known-method composition. No absolute first-ever claim is supported. The final
package should request independent validity/originality/value assessment and a
report to Scott, not imply that preparation or handoff authorizes publication.
