Saved-certificate reuse checks

All 12 known-answer cases pass. Each model/processing fixture is checked with both saved providers and with the newer future report absent. The adapter verifies hashes, environments, domains and shared maps before selecting error bounds. It reconstructs training rows and independently checks future endpoints before checking known truth.

The study uses its 36 report merges, 66 report reads and frozen support allowances in 6.21 seconds. It performs no integration, compiler invocation, optimizer start or aircraft read. Its evidence and work reservations verify.

Separate metadata tests also check the group rule: incomplete fresh coverage can use only a complete inherited range, never a partial new union. Matching identifiers alone do not establish that inherited coverage; the larger study must verify its source ancestry and coordinate frame.

Next is a separately frozen complete-cover width comparison using these saved reports. The earlier direct-only comparison remains unqualified, with all its failures preserved. Useful recording-scale ranges and observational confidence remain open.

Results · Frozen scope · Reuse policy · Roadmap.