Confidence Allocation

Most probability tools give one number: "65% yes." That number hides a lot. It does not say whether the evidence is strong and consistent, or thin and contradictory, with 65% as a rough midpoint.

Confidence allocation takes a different approach. Instead of collapsing the result to one figure, QSE preserves a full allocation across five states that describe how strongly the evidence points one way or the other. The allocation is not the same thing as a single probability. Its shape carries information that one number cannot.

QSE allocates confidence across five ordered states:

Two propositions can point in the same direction while carrying very different allocations. In one, confidence may cluster around Likely and Yes, showing a consistent positive direction. In another, confidence may be spread more broadly across Yes, Unknown and No, showing that the evidence is less settled.

QSE keeps that full shape rather than reducing it to one score. The state with the highest allocation determines the displayed state and label. Unknown remains explicit when the evidence is ambiguous, so uncertainty is represented as part of the result rather than hidden inside a single probability.

A Proposition Console result is shown in three layers. The first is the primary confidence state: the single state with the largest allocation, which determines the displayed label. The second is Directional Allocation: an interpretive aggregation of the states pointing in the same direction. For a positive direction it combines Yes and Likely, and for a negative direction it combines No and Not Likely. Unknown is not included in either directional side, because it does not point either way. The third layer is the full five-state allocation, which remains the canonical underlying distribution and is still shown separately.

This layering prevents a common misreading. A primary state of Yes around 52% can coexist with roughly 88% combined positive directional allocation, because Likely carries additional support in the same direction. At the maximum-support ceiling, roughly Yes 53% plus Likely 35% adds up to about 88% directional support. Reading only the 52 to 53% primary-state share would understate how consistently the evidence points one way.

Directional Allocation is interpretive only. It does not represent probability, odds, or any claim about an outcome, and it does not replace the full five-state allocation, which remains the canonical result. Use it to judge how one-sided the evidence is, and rely on the complete allocation for the underlying distribution.

Confidence allocation starts from a clearly stated proposition and works through the evidence in stages:

Take the proposition "A city will record more than one inch of rain next Tuesday." Normalization pins down the location, the date and the measurement source. Evidence might include the current forecast models, seasonal patterns and recent conditions.

If forecasts broadly agree on a wet system, confidence might concentrate in Likely and Yes. If forecasts disagree sharply, confidence spreads out, with meaningful weight in Unknown. Both outcomes are informative: the second tells you the question is genuinely uncertain right now.

Start with the state that has the largest allocation, because that state determines the displayed label. Then look at the adjacent states. Weight shared between Likely and Yes is a consistent positive picture with some disagreement about strength, while substantial weight in Unknown shows that ambiguity remains important.

Then read the rationale. The top factors explain why the allocation looks the way it does, and they are where you can apply your own judgment. If a key factor looks outdated or misread, you know exactly which part of the result to discount.

The public QSE methodology uses a Gaussian spread with sigma = 0.55. At maximum support for Yes, the established allocation is approximately Yes 53%, Likely 35%, Unknown 10%, Not Likely 1% and No below 1%. Yes is the displayed state because it has the largest allocation, but the complete five-state allocation remains available.

Even at maximum support, QSE does not allocate 100% to one state. Uncertainty remains distributed across adjacent states by design. At maximum support for No, the same pattern mirrors toward No, with the largest allocation at No and the remaining confidence distributed toward Not Likely, Unknown and the positive states.

Confidence allocation applies to any well-formed yes/no question: policy outcomes, sports results, economic thresholds, research questions or business decisions. Prediction markets are one place it is useful, because a market price provides an external reference to compare against. That comparison is covered in model confidence vs market probability.

The same reasoning also runs without any market at all. Proposition Console lets you enter your own yes/no proposition and receive a five-state allocation with its supporting rationale.

It is not a certainty engine. It describes how the available evidence stands at a point in time, and evidence changes. It is not financial advice. And it does not expose proprietary internals: the approach is public, while specific weights and calibration settings are not.

To go deeper, the pipeline documentation walks through each stage, the framework overview explains the design philosophy, and the glossary defines the terms used here.