[thu-nmrc]

The previous Word RAM archive registration used a standalone VERIFY.json as if it were a generic registered research package. Its inputs entry was not repository-relative, so the shared planner failed before any selected package replay; changing only that path would then make the generic validator reject this legacy package's non-generic schema.

This maintenance PR preserves the standalone capsule and removes the accidental generic registration:

• rename VERIFY.json byte-identically to STANDALONE-VERIFY.json;
• update only the standalone MANIFEST.json inventory key and its README descriptor reference;
• keep every mathematical source, data file, result, license and attribution byte unchanged;
• leave the shared planner and generic schema gates unchanged.

Validation from the exact current main base df0137efd35e11b4445734bf612bdd4c7ebc330c:

• the original current package fails its own integrity check exactly at the stale VERIFY.json inventory entry;
• the repaired standalone verifier passes its complete normal replay in 7.6 seconds (8 NTT tests, 65,537 RAM slices, 5,096 butterfly cases and 277 whole-program cases);
• its intentional Python -O rejection remains intact;
• all 32 manifest hashes pass after replay;
• the actual planner selects no generic package for this metadata-only change, while shared-control planning completes with 3,353 registered packages and excludes the legacy Word RAM capsule;
• scripts/research_package.py --changed-since HEAD^ --explain-plan ... returns an empty replay matrix.

This corrects CI registration metadata only. It does not claim a new κ, scientific admission, or a complete multiplication theorem.