Timeline
Status: history — the project arc so far.
PRISM went from a one-page proposal in mid-2024 to a decisively positive synthetic proof of concept two years later. This article records the arc in one table — including the two mid-course corrections that taught more than the clean runs did.
The arc
| when | what happened |
|---|---|
| mid-2024 | The seed: a one-page proposal — insurance-claims histories as sequences a model could learn to continue, with screening suggestions read from the continuation. |
| 2024–2025 | Concept development. The core vocabulary took shape: disjoint pools and consensus voting, the constructive-only constraint, and three-pattern learning from caught-early, caught-late, and no-signal outcomes. |
| October 2025 | The first prototype: supervised fine-tuning runs on Qwen3 0.6B and 8B base models showed that models learn the grammar of billing sequences rather than memorizing patients. Superseded, but foundational. |
| 2026 | Incorporation as a Public Benefit Corporation, making the patient-outcome priority charter-level. |
| early 2026 | The 2026 synthetic proof of concept begins: a fabricated condition injected into 25,000 synthetic patients across five pools, one independently trained model per pool. |
| 2026, waypoint | The first build selected carriers hypertension-first; the models learned "sick hypertensive patient," not the injected pattern. The build was redone with uniformly random selection. |
| 2026, waypoint | The planned preference-training secondary round ranked the right continuation but would not emit it in free generation; it was abandoned for pure SFT on the chosen rows. |
| mid-2026 | The full 16,250-run evaluation grid completes with a decisive result: fire when the pattern is present, silence when it is stripped, consensus collapsing the false positives. |
| next | Real claims data: primary and secondary training per pool, retrospective evaluation, then a first pilot — suggestions to physicians, paid only on documented early detections. |
What stayed constant
The oldest commitments are the ones that never moved. The mid-2024 proposal already framed PRISM as a recall instrument rather than a predictor — a system that may only add a screening suggestion, and whose silence must never be readable as "no need" — and that constraint has shaped every build decision since, down to banning end-of-sequence at inference. The ensemble premise is equally old: individual models were always expected to be noisy, and agreement among independently trained models was always the intended unit of evidence. Two years of building refined both ideas without weakening either; the prototype's consensus numbers are the 2024 sketch made measurable.
What changed on contact with reality
Two corrections happened mid-course, and both are recorded plainly because each was more instructive than a clean run would have been. The first was a data confound: hypertension-first carrier selection forced a full rebuild — told in the isolated signal. The second was a method reversal: the planned preference-training secondary ranked the right continuation but would not emit it, and was replaced by pure SFT — told in two training rounds.
What comes next
The synthetic prototype proves the method works when an analogous precursor pattern exists; it deliberately does not prove that real conditions have such patterns. That is the explicit leap of faith the project now tests. The next phase ingests real claims data, runs the same two training rounds per pool, and evaluates retrospectively: shown only the early portion of a real patient's record, can the ensemble surface a screening the record itself only reached years later? The retrospective read carries weight because it mirrors how the prototype operationalized earliness — firing on a history cut before the diagnosis is the same act whether the timeline is synthetic or real. If that holds, the first pilot follows — suggestions delivered to physicians through existing channels, with PRISM paid only when a suggestion leads to a documented early detection.