Logic Refinery IDR-A — Prototype
A pre-filing evidence and readiness prototype for No Surprises Act federal independent dispute resolution (IDR). This build runs on Cloudflare as a clickable, static demonstration: a human reviewer scans synthetic claim text for safety issues, extracts the four IDR decision factors, and assembles a source-cited, sealed readiness dossier. A separate Dispute Intelligence page shows clearly-labeled synthetic aggregate charts.
What this is
- A deterministic reviewer workflow: the same input always produces the same output.
- A safety scan that flags risky or out-of-scope content before anything else runs.
- Factor extraction limited to four fields: patient acuity, teaching status, market share, and prior contracted rates — each anchored to a source span.
- A sealed dossier: the reviewer sees exactly which source text supports each factor, a completeness meter, and a SHA-256 seal they can verify.
- A Dispute Intelligence page built on synthetic sample aggregates for prototype testing only.
What this is not
- It does not file disputes, and it never will from this interface.
- It makes no eligibility decisions.
- It computes no qualifying payment amount (QPA).
- It recommends no offers or payment amounts.
- It is not production-ready and makes no claim to be.
- The rules bundle is EMPTY — no rules are active until a named human reviews and approves them.
Rules status
The decision-rules bundle for this prototype is empty.
Status: awaiting-named-human-review. Nothing in this UI
applies rules; the factor cards and dossier are extraction and assembly
aids only. The live value is exposed by the Worker API at
GET /api/rules (the Reviewer page displays it when the API is configured).
How the pieces fit
The Reviewer talks to the prototype API on this site (same origin): safety scan, the Workers AI routing classifier, extraction, and dossier sealing with verification. Extraction currently runs in stub mode — the live Qwen model is wired in next.