Skip to main content

Diagnostic Game — The Consumer Head

The most speculative bet in this section. The consumer-flavored sibling of CME Delivery — same engine, different audience, different ambition.

Two heads, one body. The synthetic-case engine and LLM patient role-play that powers paid CME for licensed clinicians is the same engine that powers a consumer diagnostic-reasoning game. Build the engine for one product; ship it as two.

The hook​

A consumer diagnostic puzzle game in the spirit of the TV show House MD. The show's appeal is watching someone solve an impossible case through clinical reasoning. With LLMs we can let the player be the doctor: ask questions, get patient responses, order tests, get results, build a differential, propose a diagnosis. Solve the case = solve the disease.

The mechanics LLMs unlock that weren't possible in branching-dialogue games:

  • Dynamic patient interviews — ask any question; the patient (LLM role-played) answers in-character with medically-consistent but possibly misleading information.
  • Family interviews — talk to the spouse, the parent, the coworker; uncover history and red herrings.
  • Test ordering — order the lab, the X-ray, the biopsy. The system returns generated artifacts (CBC values, imaging report, EKG strip) that fit the case's underlying truth.
  • Differential refinement — the player builds a working diagnosis as evidence accumulates.
  • Curveballs — patients lying about symptoms, stigma-driven omissions (substance use, sexual history), unexpected lab patterns, comorbidities masking the primary issue.
  • Time pressure (for harder modes) — patient deteriorates if the wrong path is pursued.

Two product modes for two audiences​

The same engine, different framing and pricing:

AudienceFramingPricingDistribution
Licensed cliniciansEducational CME — count toward licensure requirements. Cases built from current literature.Paid subscription (see CME Delivery).Sold direct or bundled to practices.
Consumers (diagnostic enthusiasts, pre-med students, health workers in underserved areas)A puzzle game — entertainment with educational character. Not clinical advice.Free or low-cost. Ad-free if subscription; ad-supported on free tier.Mobile app stores, web. Possibly distributed via mission-aligned partners in underserved regions.

Why the consumer framing is interesting​

Three reasons, of varying weight:

1. Latent demand is enormous​

People are already using LLMs for medical questions in massive volume — billions of health-related messages per week flow through ChatGPT alone, with health questions making up around 5%+ of all messages and ~25% of weekly users asking health questions. We have quietly already crossed the threshold where humans send more medical questions to AI than to human doctors.

The consumer demand for medical reasoning is real and unsatisfied. Most current LLM-medical interactions are unstructured, often shallow, and lack feedback. A game gives the same diagnostic-reasoning practice but with structured cases, ground truth, and feedback loops.

2. Medical-education democratization​

In underserved areas — places where medical school access is limited and continuing education resources are scarce — diagnostic-reasoning practice is genuinely valuable. A health worker in a remote clinic who has practiced 200 simulated cases is meaningfully better at triage than one who hasn't, regardless of their formal credentials.

The framing here is not "this replaces medical training." The framing is "useful tool, no white-savior pretense" (the Shantaram model — operating where formal institutions don't exist or don't work, putting useful tools in people's hands and letting them figure out how to use them).

3. Shared-engine economics​

Because the engine is being built anyway for CME Delivery and for Starlight Practice's compliance-required synthetic-data program, the marginal cost of also producing a consumer product is packaging, distribution, and ad/billing surface — not engine R&D. The consumer product subsidizes itself off CME's content investment.

Liability and positioning — the only thing that can sink this​

This is where the consumer head differs sharply from the CME head, and where careful design separates a real product from a lawsuit:

  • Clear positioning every time. "This is a game / educational tool. It is not medical advice. The cases are synthetic — none of them are real patients. Decisions about your own health belong to you and a licensed medical professional."
  • No personalized triage. The game must not allow the user to enter their own symptoms and get a diagnosis. The game uses synthetic patients only. Personalized triage is a separate feature that lives inside Starlight Practice as the parent-facing 24/7 triage tool for actual patients of an actual practice — bound by HIPAA, anchored to a real chart, supervised by a real doctor.
  • Sources visible. When the game presents a teaching point, the source literature is shown. Not "AI says X" — "this case is built from this paper, here's the paper."
  • Per-jurisdiction posture. US distribution requires the strongest disclaimers; non-US distribution may have different legal posture. We are not trying to operate outside US legal review on the consumer side — we are trying to be honest about what the product is.

This aligns with Erik's general principle on AI medical advice (saved as a memory): the LLM should give its best reply, show its sources, and explicitly tell the user "I am an AI model, not a doctor. The decision rests on you, including the decision to seek a licensed medical professional."

What it looks like at MVP​

A minimum viable consumer product:

  • 25–50 cases at launch, primary-care-flavored.
  • Web-first, mobile-ready (Expo / React Native — same stack as the Starlight parent app).
  • Free tier (5 cases/month) + paid tier (unlimited + new case drops).
  • One difficulty axis (resident → attending → House-mode).
  • Achievement / streak mechanics borrowed from Erik's game-design background (GoPets, Mafia Wars).
  • A leaderboard for case-solve speed and accuracy for engaged users; explicit positioning that the leaderboard is NOT a measure of real medical skill — it's a game.

What MVP isn't​

  • It is not a chatbot.
  • It is not a triage tool.
  • It is not a substitute for medical education.
  • It does not let users role-play themselves as patients to get diagnoses.

If we keep those four "is nots" rigid, the product is defensible.

Open questions​

  • App-store policy. Apple and Google both have specific rules around medical content. Consumer medical-game content is allowed but the descriptive copy and disclaimers matter — likely a real review to navigate.
  • Liability-insurance posture for the consumer product. Even with clean positioning, expect to need cyber-and-media liability + general business liability covering the consumer product.
  • Mission-distribution partnerships. If we ever want to do the underserved-area distribution, the right path is via established global-health NGOs that already have community trust, not direct B2C in those regions.
  • Brand separation. Should the consumer product carry the Starlight brand or live under a separate brand? Probably separate brand — keeps the Starlight Practice (regulated, clinical, B2B) brand cleanly distinct from a consumer entertainment product.

Cross-references​

The diagnostic game is the most speculative bet in the entire docs site. It's here because it's a real opportunity that emerges naturally from work we are doing anyway — but it shouldn't consume engineering attention until CME Delivery has shipped and the engine is mature.