Reading guide
Status: concept — how this documentation is organized.
PRISM's documentation is a lore book: many small articles, each owning a single idea, cross-linked wherever the ideas touch. This page explains the conventions — article size, status tags, how facts are dated — and suggests reading paths for four kinds of readers.
One idea per article
Instead of a few long documents, the material is split into focused articles of a few hundred words each. Every article opens with its one idea stated plainly, develops it, and links — inline, at first mention — to whichever article owns a neighboring idea, rather than re-explaining it. Nothing depends on reading front to back: follow whatever link pulls at you and the picture assembles in any order. The sections run in the project's own logic — the problem, the data, the method, the prototype that tested it, the safeguards and business model around it, the production vision, and the history behind it — and the landing page maps every article.
Status tags
This documentation mixes locked experimental results with production intent, and the two must never blur. Every article therefore carries a status line directly under its title, built from four tags:
| tag | meaning |
|---|---|
concept | durable design principle of PRISM |
demonstrated | validated end-to-end in the 2026 synthetic prototype |
vision | production design intent; not yet built |
history | record of past work; superseded, kept for provenance |
Tags compound where warranted: "concept, demonstrated" marks a durable principle that the prototype also validated in practice. The posture throughout is honest limits — what the prototype does not prove is stated as plainly as what it does, and vision articles are never written as though the thing already exists.
Where to start
If you read only one article, read What is PRISM, the one-page overview. Past that, the table below orders a path for four common readers. The paths are suggestions, not rails; every article links onward, so any entry point works.
| audience | suggested path |
|---|---|
| Insurance / client leadership | What is PRISM → The insurance vantage point → Results-based compensation → Zero integration → Prototype results → Public Benefit Corporation |
| Clinicians | The screening gap → Clinical decision support → Constructive-only → Recall, not prediction → A GOOD example → Physician explanations |
| Engineers / researchers | The patient timeline format → Sequence completion → Pools and consensus → Two training rounds → Evaluation without a hold-out → Prototype results |
| Skeptics | Prototype overview → The isolated signal → Evaluation without a hold-out → Prototype results → A NOPE example → All models are wrong |
The skeptic's path deserves a word. The prototype was built so that doubt has specific places to land: signal isolation rules out contamination, the evaluation frame explains why there is deliberately no hold-out set, and every miss in the results is mechanically explained. If your first question is "why should I believe this," start there rather than with the overview material.
A note on time
These articles describe the project as of mid-2026. At that point the synthetic prototype is complete and its results are locked, PRISM is incorporated as a Public Benefit Corporation, and the next phase — the same method applied to real claims data — has not yet begun. The prototype's rates are an upper bound measured on deliberately clean synthetic data, not a forecast for real conditions; whether real conditions leave analogous precursor patterns is the open question the real-data phase exists to test. The timeline records how the project got here and where it points.
See also
- Landing page — the full section map
- What is PRISM — the one-page overview
- Glossary — every canon term, linked to its owning article
- Timeline — 2024 seed → 2025 first prototype → 2026 synthetic prototype → real data