Pre-read verifier implementation amendment

The primary metadata audit at registration revision d041663d completed once under its resource limits. The separately monitored pandas verifier stopped at import pandas with ModuleNotFoundError before its verify function ran or any source/ledger was read. Preserve that attempt, log and resource receipt in artifacts/multisurvey_eligibility_verification_20261012/.

No data or scientific criterion changes. Do not rerun either the primary audit or the unavailable pandas worker. Instead, freeze a distinct, dependency-free checker, research/verify_multisurvey_eligibility_stdlib.py, before execution. It independently reconstructs every source field from raw frozen CSVs, binds row-number geometry, recomputes summary counts via set intersections and verifies every retained cutout’s complete hash. It imports neither the generator nor historical research modules. It uses only Python standard-library dependencies.

The original registration and its input hashes stay unchanged, including the original verifier source. Bind the new checker, original registration and failed attempt in artifacts/multisurvey_eligibility_verification_amendment_20261012.json. Keep the ten-minute, one-CPU, 256 MiB sampled RSS and 5 GiB reserve limits. All audit, failed-check, replacement-check and report products combined remain below 1 MiB. Use a fresh guarded stage. No observations, outcome readout, fitting, threshold tuning or dependency download is authorized by this repair.

This fixes operational verification only. The audit does not establish a fresh physical-truth cohort or a successful scientific ranking. A failed replacement check ends verification with its explicit unresolved error; no automatic retries.