A decision supported by “85% confidence” sounds like a single, uniform level of certainty. Where that figure aggregates multiple underlying components, some may carry genuinely high confidence while others carry much lower confidence, blended into one headline figure that can obscure the difference.

The mechanism. Where an aggregate confidence measure combines several underlying components or sub-questions, most inputs may be highly reliable while one or two remain weak. The aggregate can still land at a respectable overall figure because the stronger inputs pull the average up. The board sees the average. It does not automatically see that the decision’s soundness may hinge disproportionately on the weakest component, which is exactly the piece the average is least equipped to flag.

Why this matters. A decision is often only as strong as its weakest material dependency, not as strong as its average dependency. Treating an aggregate confidence figure as representative of the whole decision, rather than asking which components are pulling it up and which are dragging it down, risks approving a decision on the strength of components that were never actually in question.

The test. For any decision supported by an aggregate confidence figure, ask what the components underneath it are, and whether the weakest of them is one the decision can actually tolerate being wrong about.

Evidence boundary. The general risk of aggregation masking component-level variance is well supported across risk management and organizational decision-making research; the specific framing in terms of “confidence averaging” is Obraval’s synthesis applied to this uncertainty context.