Quorvanta

Everything here is invented. There is no QUORVANTA, no soriximel tavorate, no Brennick syndrome and no Arden Quay Biosciences, and the pack holds no medical information.

The data that commercial pharma systems run on is hard to work with outside production, since approved content carries its MLR review burden wherever it goes, patient data is protected under HIPAA and prescriber data is licensed. That constrains the work long before anyone writes a test: developers end up building against stubs, nobody wants a real label or a real case record going through an external LLM or sitting in a partner’s environment, and demos get given on redacted screens that reviewers spend the meeting asking about. Quorvanta is my answer to that — an invented drug for an invented disease from an invented company, with the material a real specialty launch would generate around it, so there is something complete and unencumbered to develop against, hand to a vendor, and put on a screen.

The pack starts from a fictitious prescribing information document, and everything else derives from it with the figures agreeing wherever they appear: 105 approved claims with their references, annotated promotional and patient materials, medical information responses, a patient-support hub with case records whose benefits and prior authorization history reconcile with 223 payer transactions, serialized supply-chain events, adverse event reports, a field force with a compliance-checked call log, and a generated population at whatever volume your CRM needs. A visual identity comes with it, which matters mainly for demos, where plausible branding keeps attention on the system rather than on the placeholder styling.

For testing there are labelled scenarios, expected answers drawn verbatim from approved text, and documented traps left in on purpose — a withdrawn claim, a lot number never manufactured, a late adverse event forward — so a wrong answer is objectively wrong. Keep the answer keys out of anything the system under test can see, and the same material scores a guardrailed chatbot or an agentic patient-services assistant end to end. It’s plain JSON and XML with Python tooling, content under CC BY-SA 4.0 and scripts under GPL v3, at github.com/thowland/quorvanta.