DETAILED LOGIC, MATHEMATICS AND ARGUMENT REVIEW
Full paper and executive brief — 9 November 2026
Author: Carsten Stöcker, Spherity GmbH

OVERALL ASSESSMENT

The central argument is defensible for the stated industrial workload. No
substantive formula error was found. The review identified several argument
and cross-document clarity problems, which have been corrected in the LaTeX
and rebuilt PDFs. Original Word references remain unchanged.

The recommendation is now explicit: for industrial evidence combined from
many issuers, retained across organisations and repeatedly verified as graph
data, prefer a common W3C VCDM 2.0 profile expressed in JSON-LD and secured
with RDF-based Data Integrity as the persistent evidence format. Add an
envelope when a defined presentation, disclosure or interoperability function
justifies it. Qualified use requires a suitable eligible issuance profile.

This is an engineering recommendation under specified requirements. It is
not a theorem that JSON-LD is the only representation possible, that every
envelope is unnecessary, or that graph proofs are universally faster or safer.
Both compared architectures can carry JSON-LD. The material distinction is
the authenticated proof domain, retained verification inputs and rules binding
the application's graph view to issuer-authenticated evidence.

CORRECTIONS APPLIED

1. Separate envelope-only and dual-proof nesting.
   Locations: full §4.3, Table 4, §8.1 and Conclusion; brief §2, Figure 2,
   comparison table in §6 and conclusion; both complete-example appendices.
   The generic table heading "JSON-LD inside SD-JWT" could include the
   appendix's credential with an inner graph proof, although the table's
   properties describe envelope-only B1. The headings, caption and process
   diagram now identify B1 explicitly. The appendix's dual-proof example
   remains a separate construction, with no imported benchmark timings.

2. State the engineering choice before the regulatory consequence.
   Locations: full abstract, §8.1 and Conclusion; brief opening, §6 and §9.
   The previous conclusion focused on making a profile eligible once the
   single-pipeline objective had already been chosen. It now explains why
   the common profile is a useful default for the stated retained graph
   workload: aligning the proof domain with graph processing and reducing
   mappings and verification rules to govern. The integration benefit is
   an architectural rationale, not a measured operational saving.

3. Avoid treating JSON-LD syntax as sufficient.
   Locations: full §1.4, §8.1 and Conclusion; brief §6.
   Shared identifiers and vocabularies, controlled contexts, compatible
   validation rules, recoverable source boundaries, proof configuration and
   issuer/status checks remain necessary. T1 proves representability, not
   uniqueness or a performance advantage. Relational or other representations
   can meet abstract composition requirements with an appropriate mapping.

4. Distinguish attribution fraud from false content.
   Location: full §6.2, "Scale and automation".
   The old paragraph grouped duty evasion and counterfeit goods with attacks
   on attribution. This conflicted with the later analysis of genuinely
   issued false claims. The revision separates impersonation/forged sources
   from false classification or conformity assertions by authentic issuers.
   Provenance checks support source authentication and traceability;
   corroboration, consistency checks and risk-based examination remain
   necessary for content. The categorical automation claim was removed.

5. Allow admissible alternative assurance paths.
   Locations: brief §1 foundational-assurance paragraph and §9 conclusion.
   The earlier brief could imply that one deficient edge invalidates every
   dependent conclusion. The full model permits OR alternatives and authorised
   fallback policies. The brief now states that a required deficient link
   defeats the affected justification path; the target decision assurance
   is unavailable only if no admissible alternative satisfies the policy.
   Independent evidence remains verifiable. A common format cannot repair
   missing authority assurance.

6. Make schema-policy symmetry explicit.
   Locations: full P1 in §5.2 and Table 4.
   The witness correctly demonstrates disagreement between a schema selected
   by vct and a SHACL shape selected by RDF type. Plain VCDM credentials can
   also combine credentialSchema, RDF types and application validation rules.
   Selecting a graph-proof format does not eliminate conflicting policies.
   B1's specific obligation is coordinating vct with those rules.

7. Distinguish persistent proofs from presentation-only wrappers.
   Locations: full §8.1 and Conclusion; brief §6; both signed-example appendices.
   When a signed inner credential survives, a presentation wrapper can be
   checked at ingestion and the inner source verified directly later if
   policy allows. Two signatures need not be rechecked on every graph read.
   Reliance on outer assertions still requires their evidence and bindings.
   The complete nested example explicitly requires both proofs at acceptance.

8. Align the evidence caveat for the 1:5 ratio.
   Location: brief §1, aligned with full §8.2 and Limitations.
   The ratio remains 1 QEAA per 5 ordinary EAAs, attributed to the author's
   Spherity research estimate. The brief now also states that the supplied
   package contains no supporting dataset or methodology. It is a scenario
   estimate, not an established market measurement or a legal requirement.

9. Clarify availability of benchmark reproduction materials.
   Location: full Appendix B, aligned with Declarations and brief §7.
   The appendix previously described benchmark2/ and commands as though the
   directory accompanied the paper. It now explicitly identifies the absent
   separate archive and conditions reproduction instructions on obtaining it.
   The supplied examples/ cryptographic demonstrations are available and
   independently runnable; they are not the performance benchmark archive.

10. Repair the brief's enumeration count.
    Location: brief §2.
    "Three distinctions matter" was followed by two items. It now says two.

MATHEMATICAL AND NUMERICAL ASSESSMENT

T1 — Encoding and merge properties.
The construction is valid under its reserved disjoint namespace, injective
typed-literal encoding and stable source-scoping assumptions. Reification
handles higher arity, and shared identifiers support joins and cycles. This
is a representation result. Resource-admission rules and practical utility
do not follow from representability alone.

T2 — Cost crossover.
The difference CB(R) - CA(R) = R*delta - Delta0 is correct. For delta > 0,
A is strictly cheaper when R > Delta0/delta; the first non-negative integer
is max(0, floor(R*) + 1). The delta <= 0 case, zero-delta case, consumer-only
crossover, heterogeneous sum and common-constant mean criterion are internally
consistent. R counts additional verification events after ingestion. The
comparison assumes the same verification schedule and specified processing
paths; a local read does not necessarily require a fresh signature check.

T3 — Assurance bounds.
The lower bound, singleton upper bound, authority redirection condition and
joint-attack reasoning follow under the stated attack-feasibility and cost
assumptions. Additive costs require the specified independence/distinct-action
premise. A low-assurance credential is not automatically a feasible attack
that changes a decision. Qualification labels alone do not prove the cost
schedule, factual truth or operational safety. The policy's alternative
paths and joint attacks remain material.

C1/C2 — Qualification and cross-jurisdiction use.
The profile-eligibility requirement is conditional on the chosen homogeneous
graph-proof path and policy-required qualified attestations. Mixed-format,
source-backed and delegated alternatives remain possible. Cryptographic
attribution, issuer authority, qualified status and legal recognition are
distinct checks. Eligibility alone does not establish decision assurance.

Printed quantities.
The independent arithmetic review checked the Table 3 differences, crossover
thresholds, UC1 cost projections and denominators, UC2 daily totals, UC3
throughput/budget projections, brief/full numerical parity and the complete
example byte counts (797 compact JSON bytes and 1,534 combined-token bytes).
No substantive numerical contradiction was found.

Some rounded values do not recompute exactly from other displayed values.
Examples include 797 - 149 versus a printed delta of 649 microseconds and
a displayed 14-second worker time underlying a 55-second projection. These
are compatible with independent rounding of unrounded observations. Explicit
constructed witnesses are recorded in the arithmetic audit. They demonstrate
compatibility only: they are artificial inputs, not recovered observations.
The printed results have not been silently changed to fit rounded arithmetic.

PROOF-DOMAIN AND CREDENTIAL CHECKS

L1 separates envelope signature validity from acceptance of an interpreted
graph. Fixed authenticated contexts and processing rules address the shown
context-drift issue. Data Integrity's lossless-processing requirement supports
rejecting undefined claims that would disappear during processing.

L2 requires recoverable source snapshots, original graph positions and proof
configuration. It does not promise that isolated edges in an arbitrary merged
graph retain independently checkable signatures. Its envelope limitation is
conditional on unavailable, unreconstructible signing inputs; retained-source
and reconstructible-input alternatives remain recognised.

L3 correctly distinguishes retained-source verification, trusted delegated
results and issuer authentication of the graph representation. P2's type-scoped
context counterexample applies under the allowed disclosure profile; it does
not establish that all SD-JWT deployments are unsound. P3 and P5 distinguish
suite-specific disclosure and binding functions. P4's requirement that the
application consume the authenticated view applies to both architectures.

The final-source checks of all three JSON-LD fragments were rerun, including
safe expansion, the JSON Schema/SHACL disagreement and the real-signature
selective-disclosure witness with negative controls. The independent SHACL
check rejects the missing-unit case and accepts its positive control. The
inventory check confirms the complete signed listings and exact token match
the public fixtures and their recorded hashes. No credential bytes, keys or
cryptographic algorithms changed in this argument-review revision.

The complete-example test evidence remains: four acceptance controls and
13 expected rejections in the main runner; nine assertions and eight negative
checks in the direct Node/OpenSSL audit. The second path shares RDF processing
dependencies and is not an independent RDF implementation. The examples are
local demonstration profiles, not proof of production or qualified conformance.

PQC AND SECURITY INTERPRETATION

The full text and brief retain their distinction between credential nesting
and cryptographic hybrid signatures. Data Integrity proof sets and chains can
carry multiple proofs, but do not alone establish a secure traditional/PQC
hybrid. Validated suites, key/credential bindings, mandatory-component checks
and downgrade protection are still required. The reported benchmarks do not
measure PQC. Graph canonicalisation's resource risks and the envelope's ability
to authenticate before expansion are acknowledged; neither architecture has
an unconditional security or latency advantage.

PRIMARY SPECIFICATIONS CONSULTED

W3C VCDM 2.0, data schemas and securing mechanisms:
https://www.w3.org/TR/vc-data-model-2.0/#data-schemas
https://www.w3.org/TR/vc-data-model-2.0/#securing-mechanisms
W3C Data Integrity, securing data losslessly:
https://www.w3.org/TR/vc-data-integrity/#securing-data-losslessly
W3C EdDSA Data Integrity cryptosuites, eddsa-rdfc-2022:
https://www.w3.org/TR/vc-di-eddsa/#eddsa-rdfc-2022
RFC 9901, SD-JWT verification and processing:
https://www.rfc-editor.org/rfc/rfc9901.html#section-7.1

LIMITS AND PUBLICATION EVIDENCE

This review evaluates internal consistency, argument validity under stated
assumptions, printed arithmetic, credential examples and document production.
It is not external peer review, a production-security certification, a legal
opinion or independent reproduction of the original performance measurements.
The absent benchmark implementation/run archive remains the main reproducibility
gap. The 1:5 scenario estimate remains unsupported by supplied empirical data.
Those limitations are disclosed in both documents; neither is disguised by
the successful cryptographic example tests or the arithmetic consistency check.

The accompanying changes JSON records exact before/after passages. The formula
audit supplies its analysis and executable arithmetic checks. The separate
conversion-qa.txt records final build, reference, licence and visual checks.
