Runs the atmoq relay-conformance corpora through the real decode and validation path and writes one JSONL verdict per case, so the table rows are measured rather than inferred. The atmoq README already documented this entry point; it had never been committed here. The harness reports the relay-level verdict (would zlay drop this frame?) rather than the raw result of a validation function, mirroring the drop sequence in frame_worker.processFrame. Prior repo state for the stateful sync-1.1 sequences is held in memory so no database is needed. Measured against the published baseline: 43 of 49 rows unchanged, the five intended reject flips confirmed, and one regression surfaced — sync11/commit2-valid, a valid @atproto second commit, is dropped with PrevDataMismatch. See the report accompanying the zat investigation.