Why we say 'forecast,' not 'predict'
Voice discipline as substance — the word is the claim.
A "prediction" is a single outcome, asserted. A "forecast" is a distribution over outcomes, with its own doubt carried in its shape. The word we chose is the claim we are willing to defend, and the word we rejected is the claim no honest system can keep.
There is a word we do not use. "Predict" implies a single outcome the speaker claims to know. A forecast is a distribution over outcomes, with the doubt carried in the shape. The difference is not stylistic; it is the difference between a bet you cannot hedge and a bet that tells you how much to hedge. We chose the second word because it is the one the architecture can actually back.
The architecture is a distribution, not a number
The Oracle does not output a point estimate. The sisters — five personalities reasoning over the same versioned GNL world model — each produce a draft, and deino-oracle merges them into a normalized Ensemble: probabilities sum to one, scenarios sorted descending, entropy in nats. Normalization lives in exactly one place (INV-1) so every consumer can rely on the invariant. The deliverable is a shape — a cone — not a number. Calling that shape a "prediction" would be lying about what the system produces.
Why the banned word matters in practice
A regulator reads "we predict X" as a truth claim — and then asks the question that kills it: "do you certify this is correct?" No honest system answers yes. A regulator reads "we forecast a distribution with this entropy and these modes" as a provenance claim — and asks "can you reproduce how you got this?" That is the question the manifest answers, by construction. The word chooses the question. We chose the word that invites the question we can answer.
"Predict" invites the question no honest system can answer. "Forecast" invites the question the manifest answers by construction. The word chooses the question.
- A prediction is a point; a forecast is a distribution. The architecture produces the second, not the first.
- A prediction is a truth claim; a forecast is a provenance claim. Only one survives an examiner.
- The word is the claim — and the claim has to be one the substrate can back (Certified ≠ True).
The one exception, and why it is a critique
There is exactly one place where "point predictions are a liability" appears — and it appears as a critique, not a capability. Point predictions are a liability because they collapse the doubt into a number and leave the CRO with no hedge. That is an argument against the thing, not a description of what we do. We forecast cones; we do not predict points. We are pre-revenue, building the POC; the cones are produced, and the calibration that earns the word "probability" is gated behind Gate 1b. The voice discipline is not a marketing preference — it is the honesty label on what the system actually delivers.
The word we chose is the claim we can defend: a distribution, reproducible, with its doubt visible. The word we rejected is the claim no honest system can keep: a single outcome, asserted as known. Voice discipline is not aesthetics; it is the contract.
— voice note