'What if you are wrong?' — the question a distribution answers
The board question that kills a point estimate, answered by a shape.
The board question that ends every point-estimate forecast — "what if you are wrong?" — has no answer when the forecast is a number. It has a full answer when the forecast is a distribution: the tails are the answer, the entropy is the size of the answer, and the modes are the contingency. The question does not kill the forecast; it kills the point.
Every CRO who has brought a point estimate to a board has met the same question: "what if you are wrong?" It is the question that ends the conversation, because a point estimate has no answer. "4.2%" is either right or it is not, and "what if it is not" is an accusation with no reply. The board is not being difficult; the point estimate is being dishonest — it presented a number without the doubt the number rests on, and the board is asking for the doubt back.
The distribution answers before the question is asked
A probability distribution answers "what if you are wrong" before the board has to ask. The tails of the distribution are the "wrong" scenarios, with probabilities attached. The entropy is the size of the doubt. The modes are the contingency regimes. When the board asks the question, the CRO points at the shape: "if the tail regime fires, here is the scenario and its probability; we are hedged for it at this weight." The question becomes a contingency, not an accusation — because the doubt was already in the forecast, not bolted on in the meeting.
The point estimate is the liability
This is why point predictions are a liability, not a feature. A point estimate is a distribution the model collapsed, with the doubt discarded. The board then spends the meeting reconstructing the doubt the model threw away — asking "what about 5%, what about 3%, what if the hike does not happen" — and the CRO has nothing to point at. The distribution keeps the doubt in the forecast, where it belongs, so the board spends the meeting on the decision, not on reconstructing the doubt the model collapsed.
A point estimate is a distribution the model collapsed, with the doubt discarded. The board then spends the meeting reconstructing the doubt the model threw away. The distribution keeps it where it belongs.
- The tails are the "wrong" scenarios — with probabilities, not hand-waving.
- Entropy is the size of the doubt — the honesty field a CRO reads before the scenarios.
- The modes are the contingency — the regimes the model cannot distinguish between.
Calibration is what makes the answer trustworthy
A distribution answers "what if you are wrong" only if the probabilities on the tails are calibrated — if a 10% tail actually happens about 10% of the time. Without calibration, the shape is a more sophisticated guess than a point, not a better one. The calibration loop (deino-eval, Brier and log-loss, shadow runs against a realized timeline) is what makes the tails worth hedging. Gate 1b — beat a disclosed baseline on one series — is the gate that lets the shape be positioned as accurate. We are pre-revenue, building the POC; the shapes are produced, and the calibration that makes them trustworthy is gated. The question does not kill the forecast; it kills the point. The shape is the survival.
The board question "what if you are wrong?" is not an accusation when the forecast is a distribution; it is a pointer to the tails. The distribution answers it before it is asked — and that is why the point estimate does not survive the room.
— board-prep note