The review is intentionally divided between roles. A reviewer may fill more than one role only if that overlap is disclosed.
Read the frozen paper independently, then use the paper inventory and author
review cards. Confirm the mathematical objects, normalization of valuation
and quadratic defect, source/target orientation, quantifier order, strict
inequalities, rank conventions, exceptional cases, and one-based to zero-based
index translation. Record disagreements rather than silently repairing them.
For Beli 2020, explicitly decide whether the r_1 coefficient printed in
Theorem 3.1(3.2.1--2) agrees with the 2r_1 coefficient obtained by direct
substitution into Theorem 2.1, or should be recorded as a source correction.
For the He--Hu and He ADC entries, use only the publisher version
of record named and hashed in papers/<paper-id>/paper.json; the listed arXiv
copies are comparison sources and cannot silently amend the published text.
Build a clean clone, inspect the actual declaration signatures rather than
their names or comments, expand DyadicContext and the representation
definitions, run the axiom audits, and check the import graph for circularity
or an imported equivalent of a target theorem. A reviewer may instead download
one paper's Review Kit from papers/INDEX.md; its manifest records both covered
and explicitly excluded scope.
Each signed review must identify:
- paper version and SHA-256;
- repository full commit and tag, if any;
- reviewer name, affiliation or public profile, role, and date;
- exact files or theorem cards reviewed;
- build platform and command results for a reproducibility review;
- one decision for every core review card;
- all reservations, exclusions, and conflicts of interest.
Signatures belong in docs/audit/IndependentReviewSignoff.md or in an
immutable linked issue/review. A name added by the project author or an AI
agent is not independent sign-off.
The project may be promoted from Grade B to Grade A only after every core
result is VERIFIED_MATCH, the clean-clone build is independently repeated,
and no unresolved core semantic issue remains.