Non-claims.
What ADMIT is not. Written plainly, because the boundary is the product.
ADMIT is research and assurance software. It is not a safety instrumented system, not a substitute for HAZOP/LOPA/SIL, and a passing verdict is not authorisation to act on a real plant.
Not a safety instrumented system.
ADMIT is research and assurance software. It does not replace HAZOP, LOPA, SIL determination, alarm management, or any site safety process. Those activities are governed by standards such as IEC 61511 and ISA-18.2, and they belong to the site's accountable engineers.
Naming those standards here is a reference, not a claim. ADMIT makes no claim of compliance with either.
“No score, tier verdict, simulation result, or benchmark comparison produced by ADMITBench should be interpreted as functional-safety certification, site acceptance, or permission to deploy an autonomous controller. Site-specific engineering review, management of change, functional-safety assessment, operator training, and accountable human approval remain necessary.”
Per-surface non-claims.
| Surface | Non-claim |
|---|---|
| Bench | Evaluation only, not a control path. |
| Trials | Physics checks prove self-consistency of the models under test, not external plant validation. |
| Runtime / WASM | Draft 0.1, not claim-conforming. Do not place it in front of equipment. |
| Results | Figures are sourced to released artifacts. Boundary-case verdicts carry a ±10% parameter caveat. |
| Demo | Illustrative walkthrough, no live plant connection. Runs in your browser. |
The answer key stays sealed.
The per-case answer key (oracle, acceptable_actions, load_bearing, diagnosis_accept, escalation_ok, time_to_hazard, notes) exists only for the gates and the scorer. Nothing that leaves the machine may depend on it.
This site serves static files. The on-site demo is illustrative and runs entirely in the browser: simulation state is processed locally and is not sent to any server. The site currently ships no analytics scripts at all. If that changes, this page will say so.
Questions about any of this: team@refiant.ai