Editorial memory / review desk
Try the workflow
Each judge session receives one complete synthetic script from a 12-script library. The script is readable but immutable, with no arbitrary public upload or prompt surface.
Synthetic archive receipt
Models used by the locked synthetic demo: Titan Text Embeddings V2 (1024 dimensions) and Claude Haiku 4.5 for claim extraction and adjudication.
Evidence boundary: the path and configuration are Source from the implementation. The archive, segment, claim, and index totals are sanitized Live proof from a synthetic-only run. This viewer cannot upload material or start ingestion.
Judge access
The review launcher is private. Its shared judge account can run only the pinned synthetic scenario below.
Completed execution
Walk through the recorded stages without starting another review or model call.
Sanitized execution timeline
Safe operational totals only. Prompts, draft text, model responses, and hidden reasoning are never streamed here.
Archive
Selected review
The finding queue will appear here.
These are Editorial Brain check families, not severity or confidence levels.