Recall, not prediction
Status: concept, demonstrated — enforced mechanically throughout the prototype.
PRISM is a recall instrument, never a predictor: its only possible statement is "consider this test," and it is built at every layer so that its silence can never be read as a judgment about a patient.
The distinction that shapes everything
A predictor assigns a calibrated probability, and both ends of that scale carry meaning — including the low end. The moment a screening system speaks that way, its silence acquires a use: "the model saw no need, so the test need not be covered." A tool built to catch missed care would become, in the same breath, an instrument for denying it. PRISM refuses that failure mode structurally rather than by policy. It has no vocabulary for this patient is fine, and nothing in its output can be acted on as a denial.
This constraint has governed the design since the project's original one-page proposal, and it explains several build choices that look strange until you see what they defend against.
EOS is banned; continuation is forced
At inference, every model in the ensemble is denied its end-of-sequence token and forced to keep continuing the six-column table for a fixed budget — this round, roughly 1,500 tokens, ten or more complete rows. A model allowed to stop would be making a statement: ending the table reads as "nothing more is coming," which is exactly the calibrated no-need prediction PRISM must not emit. Banning the token removes the option entirely. The forced budget is a convenience for measurement — every run produces same-shaped output — but the deeper reason is that stopping must never be available to interpret.
The read is binary. A continuation either contains the TEST code or it does not — a string match, not a judgment call. No confidence score attaches to a single continuation, and none is invented downstream.
Asymmetric output
The two possible outcomes of a run are not mirror images, and the system treats them accordingly:
| outcome | what it means | what happens to it |
|---|---|---|
| the TEST appears in the continuation (a fire) | "consider this test" — a pattern worth a human look | counted toward consensus; may become a suggestion to a physician |
| the TEST does not appear (silence) | nothing | never stored, scored, or reported as any statement about the patient — not "low risk," not "cleared" |
Only one of these is an output at all. A fire is a flag; silence is the absence of a flag, and PRISM keeps no ledger of it. There is no "screened and negative" record anywhere in the system, because the system cannot produce that claim in the first place.
Silence measured, and found empty
This principle is usually asserted on ethical grounds. The 2026 synthetic proof of concept was able to measure it. Across the full evaluation grid, every miss on a signal-bearing patient traced to one mechanical cause — a model fires if and only if its prompt contains at least one precursor row — so silence reflected missing input, never a low-risk read of the patient. The full accounting, miss by miss, is in the prototype results.
That finding makes denial-by-silence statistically indefensible, not merely disallowed: a non-fire carries information about what the prompt contained and none about what the patient needs. And the synthetic world is the favorable case — real claims histories are less complete, so on real data silence can only mean less.
A flag, not a probability
What PRISM ultimately reports is consensus: how many independently trained models surfaced the same test for the same patient. That count is a flag — "N of the ensemble agree this is worth a look" — and it is deliberately never converted into a percentage chance that a patient does or does not need care. Consensus filters noise, as pools and consensus explains; it does not become the calibrated scale the recall posture forbids. A flag invites attention; a probability invites arithmetic against a threshold, and thresholds are where silence gets weaponized.
Everything else points the same way
Two other layers of the system reinforce this stance rather than merely coexist with it:
| layer | how it reinforces recall |
|---|---|
| constructive-only architecture | the pipeline counts positive suggestions only; no component can emit "don't test" — denial is unrepresentable, not merely suppressed |
| results-based compensation | PRISM is paid only when a suggestion leads to a documented early detection — it earns by catching what was missed, and no revenue path runs through gatekeeping |
One makes denial impossible to build; the other makes it unprofitable to want. Together with this article's operational stance — a banned token, an asymmetric record, a count that stays a count — the result is a system that can speak only to suggest, and whose silence belongs to no one.
See also
- Constructive-only architecture — the structural half of the same guarantee
- Pools and consensus — where the N-of-M flag comes from and why it filters noise
- Prototype results — the firing grid, every miss explained, and the honest limits
- Results-based compensation — the business model that only pays for catches