Historic Sql Patterns

It reads sharded pattern input files, verifies every join column through entity_details, and emits typed evidence records with slugs, narratives, and constituent template IDs. The workflow is strict: read the shard, identify durable cross-table intents, emit one pattern per intent, then stop. It won't write wikis or semantic layer sources itself, just produces the pattern evidence for downstream projection. The identifier verification protocol is thorough, requiring entity_details checks and sometimes SQL probes to confirm literals exist before committing anything to output. Good for mining institutional analytical knowledge from logs without inventing tables that don't exist.