Use fictional or redacted material.
Do not enter names, dates of birth, addresses, account or case numbers, medical record numbers, patient information, confidential records, regulated records, or other identifying information.
PRVLocal-only preserved-research lab boundaryno network evaluation call
Keep the case yours.
The public lab runs an earlier preserved-research evaluator in your browser memory. It is not the current R1 runtime. It does not send a case to Project Shadow, persist it in browser storage, or call a model. You control every copy and download.
Deterministic and in memory.
The earlier 28-gauge preserved-research evaluator makes no network request and uses no localStorage, database, or file upload. It is not the current R1 runtime. Resetting or leaving destroys the current case.
You move the packet.
The comparison desk creates a prompt you may copy into a model of your choice. It validates only JSON you paste back. Your chosen model has its own data-handling terms; Project Shadow does not control them.
Narrative is excluded by default.
Markdown and JSON exports are generated locally. Case narrative enters the export only when you explicitly select that option. The reference hash is an integrity aid, not a signature.
No output authorizes action.
A beta reference receipt is not a canonical runtime receipt, validation, certification, legal or medical advice, or deployment authorization. A model output cannot lower the deterministic floor.
Reference beta means challengeable.
Built-in checks and use of this local research surface support continued evaluation. They do not establish independent validation. Report reproducible errors through the source repository once its public boundary is complete.