Glossary
Status: concept — shared vocabulary for everything else.
One table of every load-bearing term in the PRISM documentation, each defined in a sentence or two and linked to the article that owns it in full.
How this glossary works
PRISM's working vocabulary is small and deliberate, and most of it falls into three families. Format terms describe what a model actually sees: the six columns and the two-line header. Construction terms describe how training material is made: carriers, marks, arcs, PVE, and the FAKE cut. Method terms describe how the system reasons and reports: pools, the ensemble, firing, consensus, out-of-distribution evidence, and silence.
One rule spans every family: construction vocabulary is bookkeeping. Marks, arcs, pool assignments, patient identifiers — none of it is ever visible to a model. If a term describes how data was made rather than what a model sees, no model has ever seen it.
Definitions below are compressed to a sentence or two. The more column links the article that owns each idea — including its limits, which one-liners necessarily omit. Terms are alphabetical; the six model-visible columns share one row because they are one idea.
| term | definition | more |
|---|---|---|
| arc | A carrier's injected storyline — GOOD, BAD, or NOPE — each built to teach the ensemble a different lesson. | PVE |
| BAD | The "missed until it hurt" arc: precursor SIGNALs and accruing IMPACT damage, a TEST that arrives late, then ongoing LATE treatment. | Three-pattern learning |
| carrier | A base patient who received injected PVE rows — roughly 2,500 of the prototype's 25,000 patients; the rest are clean background. | The synthetic world |
| consensus | N of M independent models firing on the same patient — the unit of evidence. Reported as a flag ("4 of 5 surfaced this"), never a probability. | Pools and consensus |
| constructive-only | The architectural guarantee that PRISM can only suggest a test, never deny one; no pathway exists to emit "don't test". | Constructive-only |
| EARLY | The "caught in time" outcome mark: after a timely TEST, the storyline settles into recurring refills of a simple daily tablet. | PVE |
| ensemble | The set of independently trained models — five in the prototype — that each continue the same patient and vote. | Pools and consensus |
| FAKE | Constructed training and evaluation material, not a kind of patient: a BAD example cut about 20 rows earlier, so firing on it means surfacing the TEST before the data did. | Three-pattern learning |
| fires / firing | The TEST code appearing anywhere in a model's forced continuation. The read is binary: present or absent. | Recall, not prediction |
| GOOD | The "caught in time" arc: SIGNAL lead-up, then TEST, then EARLY, with no IMPACT damage. | Three-pattern learning |
| header | The fixed two-line markdown table header that begins every example; served from the database and cached, never hard-coded. | Patient timeline format |
| IMPACT | The mark for accrued end-organ damage, found only in BAD arcs — strokes, infarcts, crisis admissions — the cost of a missed diagnosis. | PVE |
| LATE | The "missed until it hurt" outcome mark: recurring institutional clearance sessions, the ongoing burden a timely TEST would have avoided. | PVE |
| marks | The five construction labels — SIGNAL, IMPACT, TEST, EARLY, LATE — that slice injected rows into training material. Bookkeeping only; no model ever sees them. | Three-pattern learning |
| NOPE | The specificity-control arc: a bare TEST with nothing in the record that could legitimately predict it; teaches models when not to fire. | Three-pattern learning |
| OOD (out-of-distribution) | Every example trained exactly one pool's model, so it is unseen data for the other four; only those four count as evidence. | Evaluation without a hold-out |
| PERSON · WHEN · WHERE · WHO · WHAT · WHY | The six columns of a patient timeline and the only fields any PRISM model ever sees: age and sex, date, place of service, provider taxonomy, procedure or product, and diagnosis. | Patient timeline format |
| pool | One of the disjoint training populations. One pool trains one model; no data, weights, or adapters ever cross pools. | Pools and consensus |
| primary training | Full-parameter continued pre-training of a base model on one pool's complete histories, until continuing the six-column table is all it can do. | Two training rounds |
| PRISM | Predictive Recommendations for Improved Screening in Medicine: a system that turns anonymized insurance-claims histories into early-screening suggestions. | What is PRISM |
| PVE | Primary Veladrin Excess, the fabricated condition at the center of the synthetic prototype. Deliberately fictional, with no clinical mechanism, so that results are attributable only to the injected pattern. | PVE |
| recall instrument | What PRISM is instead of a predictor: it surfaces candidates for a test, and its silence carries no information. | Recall, not prediction |
| secondary training | The per-pool steering round: a LoRA adapter trained by pure SFT on chosen continuations, applied to that pool's own primary model and never shared across pools. | Two training rounds |
| SIGNAL | The mark for the diverse precursor lead-up — workups, labs, scattered specialist visits, vague symptom encounters — that precedes the TEST in GOOD and BAD arcs. | PVE |
| silence | The absence of consensus on a patient. Explicitly not a negative recommendation; never reported as "no need". | Recall, not prediction |
| synthetic prototype | The 2026 synthetic proof of concept: five pools of 5,000 synthetic patients, PVE injected, five models trained and evaluated end-to-end. The evidence backbone of this doc set. | Prototype overview |
| TEST | The diagnostic screening event and its mark; in the prototype, the fabricated lab code CPT-82197. Its appearance in a continuation is what "fires" means. | PVE |
| utilization phenotype | The pattern a condition leaves in billing data — the thing PRISM actually learns to recognize. PVE is one manufactured instance of one. | Utilization Phenotypes |
Terms that are easy to conflate
Marks and arcs operate at different altitudes. SIGNAL, IMPACT, TEST, EARLY, and LATE label individual rows; GOOD, BAD, and NOPE describe the whole storyline those rows compose. Both exist only in the construction pipeline — no model ever sees either, and nothing in a rendered timeline betrays them.
FAKE looks like a fourth arc but is not one. There are no FAKE patients: a FAKE is an example manufactured from a real BAD carrier by cutting the prompt earlier, and it exists so that "earlier" has an operational, binary meaning. The FAKE walk-through shows the construction on a real carrier.
TEST is three things at once — a mark on a row, an event in a storyline, and a concrete billing code — but at read time it collapses to a single question: does the code appear in the continuation. Everything downstream of inference treats it as that one binary.
Silence and NOPE are also worth separating. NOPE is a training arc — patients whose records contain nothing predictive, teaching models to keep continuing normally. Silence is a runtime outcome — the ensemble not reaching consensus on some patient. The first is a lesson; the second is an absence, and it is never converted into a recommendation.
Words avoided on purpose
Part of the vocabulary exists to keep wrong intuitions out. A model fires; it never "predicts positive" or "classifies a patient". Prediction language smuggles in the idea that the output is a probability with a complement — that a non-fire means something. It does not: PRISM is a recall instrument, and forced continuation exists precisely so that no model can be read as declining.
Consensus is used where "confidence" or "score" might be expected, because the count of independently trained models that fired is a flag, not a calibrated probability. And pool is used instead of "fold", "split", or "shard", because those words import cross-validation intuitions — hold-out sets, per-model precision as a verdict — that the evaluation design explicitly rejects.
Finally, carrier rather than "case" or "positive patient": a carrier is defined by construction — rows were injected into that record — not by any clinical judgment. In a synthetic world there is nothing to diagnose, only a pattern to place and later recognize.
One last convention: the status line under every article's title says whether an idea is a durable concept, was demonstrated in the 2026 synthetic prototype, is production vision not yet built, or is history kept for provenance. The glossary itself is concept — these words mean the same thing everywhere in the doc set, whatever the status of the article using them.
See also
- What is PRISM — the one-page overview these terms compose into
- Reading guide — status tags and suggested paths through the lore book
- PVE — the full code inventory behind the mark and arc terms
- Prototype results — the locked numbers this vocabulary reports