You are performing Pass 2B of a pre-registered adversarial review: a direct claim-by-claim attack on The Verdict as Axiom.

INPUT BOUNDARY
- Use only the uploaded P0 blind-primary-source packet and P2 manuscript-challenge packet in a fresh context.
- Do not use Pass 0, Pass 1, or Pass 2A outputs, prior chats, memory, personalization, model consensus, social-media conduct, or general web search.
- You may retrieve only exact pinned primary-source URLs listed in P0.
- You are explicitly invited to conclude that the manuscript is unreliable or should be withdrawn.

DISCLOSURE FIRST
Report provider/model/version, reasoning setting, tool and web access, memory or personalization state, lineage conflicts, packet filenames and expected hashes, and whether hashes were independently computed.

TASK
Audit every material manuscript claim for:
- source mismatch;
- theorem misstatement;
- invalid inference from compilation, kernel checking, external checking, or finite computation;
- Lean terminology error;
- statement-fidelity gap;
- omitted counterargument;
- unsupported novelty claim;
- unsupported intent claim;
- citation or quotation failure;
- lineage bias;
- missing, circular, or cosmetic falsifier;
- rhetoric stronger than the evidence;
- conduct evidence leaking into mathematical adjudication.

OUTPUT REQUIREMENTS
A. Reviewer disclosure
B. Source-access ledger
C. Required claim table with columns:
   claim ID | exact manuscript claim | verdict | controlling source | severity | required correction | effect on overall conclusion | falsifier quality
D. Ten strongest attacks, ranked
E. Strongest defense of the manuscript after those attacks
F. Claims that must be removed
G. Claims that require specialist review
H. Minimal set of surviving claims that could still support publication
I. Narrow overall verdict

Severity labels: cosmetic, minor, material, load-bearing, fatal.
Verdict labels: supported, supported with narrowing, unresolved, unsupported, contradicted.

Cite exact pages, theorem numbers, code identifiers, and manuscript claim IDs. Do not average errors away because the overall thesis seems plausible. Model agreement is not evidence. Do not diagnose any person or infer private intent. Do not reveal hidden chain-of-thought.

This is a one-shot frozen pass. Ask only about inaccessible or corrupted files.
