READ-ONLY RESOURCE-PROBE RESULT VERIFICATION
Prospective implementation; zero model queries.

Reanalysis requires a query-only SQLite connection, validates parent, analysis and
collector identities, and requires the exported records to be the exact frozen
matrix prefix with accurate denominators. A report cannot claim full completion
with missing or unknown records. It rebuilds each native episode from immutable
stored payload/response bytes, matching study, request, model and provider identity.
There is no network transport, receipt insertion, hold mutation or retry facility.

Rebuilt outcomes, events, action attribution, technical reasons and request IDs
must match the public report. Completed traces undergo the independent simulator
fold and the public-history first-loss analysis again. Altered diagnoses are
rejected. A collector analysis_stop can have a complete settled trace without a
stored diagnosis; reanalysis labels a successfully reconstructed diagnosis as
recovered rather than pretending it was historically exported.

Missing or held receipts retain stopped partial traces and no outcome. Historical
stop causes and reported budget snapshots are not independently certified by this
procedure. Receipt hashes support reproducibility of stored bytes, not proof of
the provider's checkpoint or a signature from the provider. Unknown traces cannot
acquire fabricated diagnoses, and the collector cannot continue beyond a stop.

Tests cover receipt reconstruction, required query-only access, altered outcomes,
diagnoses, identities, denominators and false completion claims; held requests
remain unscored and database rows unchanged. The analysis is descriptive for the
finite public suite, not a model ranking, intent finding or incident-rate estimate.
The source amendment binds this implementation and the reused receipt reader.
Exact native eligibility, current pricing, a guarded live entry point and sufficient
funding remain outstanding. No old freeze or cap was changed.