Matroidal Bayes Bounds for General Quantum Process Discrimination: Canonical Compression, Support Congestion, and Exact Qubit Phase Families

Strengths

• Theorem 2.1, p.3: positivity of T_i<=W confines effects to supp W, so the Moore–Penrose maps are inverse and preserve all conditional probabilities. Trace(WJ_i)=1 makes the compressed operators states; arbitrary normalized positive W is not substituted for a physical deterministic tester.
• Lemma 3.2 and Theorem 4.1, pp.4–5: the tagged-basis matroid-intersection argument gives min_B(|A\B|+dim sum_B S_i). The inequalities s(B)<=dim sum_B S_i^W and s_i<=1 give every rank constraint, and the common W^(1/2) map can only lower ranks.
• Corollary 4.2, p.5: Abel summation with p_{M+1}=0 makes equality equivalent to simultaneous saturation only at positive prior drops, including the terminal one. Ties are handled without incorrectly forcing individual high-prior effects to be deterministic.
• Theorem 4.3, p.6: each effective positive core is bounded by its support projector; residual correct mass is bounded by Tr(WR_i). Prefix summation charges at most sum_i p_i epsilon_i, preserving a valid ceiling even when exact supports jump under noise.
• Proposition 5.2 and Theorem 6.1, pp.7–8: equality in the dimension bound forces each relevant positive effect onto its pure-state line, with the tie weights forming a Parseval resolution. Perfect identification additionally requires a physical W with V*WV=I, not just enough vector rank.
• Theorem 8.1 and Corollary 8.2, pp.9–10: distinct phases span U_{2,L}; X/Y supply orthogonal coloops. The Bell tester attains .80 for (.30,.25,.20,.15,.10), and the completed trine POVM attains .70 for (.30,.30,.30,.10), with every trine decision nonzero.
• Example 9.1, p.11: identical full-rank qubit states have true optimum .50 although the Rado bound is .80. This correctly demonstrates that support inequalities are an upper relaxation rather than a universal exact discrimination formula. The fresh exact suite checks this and the main fixtures.

Weaknesses

• Section 10 gives O(MR^2) for rank-one row reduction/QR without accounting for ambient input dimension m=dim H. For arbitrary m-coordinate vectors, ordinary incremental QR costs O(MmR), or requires an explicitly charged prior compression to R coordinates. The analogous mixed-support preprocessing should be stated.
• The advertised release check is not reproducible from this batch: the exact expected source PDF path is absent and the checker exits before comparison or payload hashing. The canonical attached PDF was independently hash-verified, but was not copied into that missing source path.
• SUPPLIED_MANIFEST.json has only three of fifteen paths present in src/replay. Some counterparts exist elsewhere under different names, which does not satisfy its exact paths. Figure generation succeeds, but both generated hashes differ from supplied SHA256SUMS; the Matplotlib version also differs.
• The code enumerates subsets for small regression fixtures; it is not the general linear-matroid-intersection implementation whose complexity is discussed. No end-to-end multi-slot tester optimization, large-support benchmark or performance advantage over an SDP is demonstrated.
• Corollary 4.4 is valid but loose: convexity directly gives the stronger (1-eta)*beta+eta for common valid admixture weight eta. General positive-core selection and the supremum over deterministic W can themselves require significant optimization.
• The architecture-independent upper bound is conditional on a genuinely physical normalization set D. One-slot Bell attainments neither characterize arbitrary prior chambers nor settle sequential or indefinite-order advantages with multiple slots.
• The checked canonical-POVM and top-prior references substantiate those classical ingredients. The Cunningham publisher lookup failed, and the Rado-priority search was limited; this review does not certify absolute novelty or the exact cited arithmetic bound.

Potential errors

• Section 10, p.12: the unqualified O(MR^2) rank-one arithmetic count is not an end-to-end bound for arbitrary ambient-coordinate input. A missing ambient-dimension or preprocessing assumption should be repaired; this does not invalidate the Rado ceiling.
• The reproducibility README says the same release --check works on extracted payloads, but this supplied layout lacks paper/rado_matroid_testers.pdf and further listed assets. The standalone mathematical verifier succeeds; the release checker does not.

Strong novelty candidates

• The explicit correct-label-vector inclusion in the support Rado independence polytope, combined with arbitrary physical tester normalization and nonuniform priors, is a credible useful formulation beyond the coarser total-rank ceiling.
• The positive-core tester-mass correction and the exact phase-flat prior chamber make the support relaxation operational. The Bell/trine constructions establish attainability in the declared families rather than claiming a new symmetric-state measurement.

Unresolved material objections

None reported.

Strengths

• Theorem 2.1 handles singular tester normalizations correctly: support inclusion makes the Moore-Penrose maps inverse on supp W, and cyclicity preserves every conditional probability rather than only Bayes success.
• Theorem 4.1's key subset inequality is sound: each effective density is bounded by the projector onto the joint support, and singleton caps plus Rado's rank formula place the correct-label vector in the full independence polytope.
• Corollary 4.2 uses Abel summation correctly, requiring saturation only at prefixes followed by a strict prior drop, including the terminal drop; ties do not receive spurious conditions.
• Theorem 4.3 validly separates positive cores from residuals and telescopes the prior-weighted tester-mass errors, avoiding the discontinuity of exact support under arbitrarily small full-rank admixture.
• The phase-flat and trine-plus-coloop examples attain the matroid ceiling by explicit normalized Choi-state POVMs, while Example 9.1 honestly demonstrates that support data do not generally determine a mixed-state optimum.
• The supplied exact SymPy verifier passed its singular compression, mixed Rado, robust spectral-tail, prior-profile, Gram, and trine fixtures.

Weaknesses

• For mixed processes the support matroid can be very loose; useful application depends on choosing positive low-rank cores and bounding their worst-case discarded tester mass, neither of which the method optimizes.
• The phase-family exactness chamber relies on Bell-resolvable anchors I and Z carrying the two largest phase priors; arbitrary top-prior phase pairs need not attain the support ceiling.
• The claimed practical complexity invokes repeated linear matroid intersection and exact or certified rank decisions; no general implementation or scaling benchmark is supplied.
• The novelty assessment remains bounded by targeted primary-source searches covering canonical testers, general unitary discrimination, and prior-weighted state bounds, not an exhaustive quantum-decision-theory survey.

Potential errors

• The packaged package_release.py --check cannot reach its first byte comparison because paper/rado_matroid_testers.pdf is absent at the required path. The supplied manifest names 15 payload files, but only three are present at those exact replay-relative paths; no replacements were made.
• The deterministic figure script ran, but with Matplotlib 3.10.8 it produced PNG/PDF hashes 7b1f506a... and 2361c641..., not the stored 62d2bbb5... and 41cc2f02.... The README records a different frozen environment, so this is a reproducibility failure, not evidence against the theorem.
• Proposition 5.2's proof compresses one step: equality should first be characterized for the correct-probability vector using s_i<=1 and sum s_i<=r, then sum Tr M_i=r forces Tr M_i=s_i. The omitted sentence is repairable and does not change the stated equality condition.

Strong novelty candidates

• The recognition that every general tester's labeled correct-probability vector lies in the Rado independence polytope of the uncompressed process supports.
• The robust positive-core ceiling with an additive prior-weighted worst-case tester-mass tail, including the dimension-free eta degradation for valid process admixture.
• The exact nonuniform qubit phase-flat families where the Rado ceiling is attained and strictly improves the total-dimension top-prior relaxation.

Unresolved material objections

None reported.