www.lesswrong.com/posts/ybYBCK9D7MZCcdArB/how-to-measure-anything#Step_5__Make_a...

In the example just above this sentence, the variable being sliced is units sold, and the decision threshold is 200,000 units because the campaign costs $5M and earns $25 per unit. Multiplying a segment midpoint by its probability gives an expected number of units, not an expected opportunity loss in dollars.

Hubbard’s own exhibit for this section says the opposite: for each increment below the threshold, EOL is computed as a small probability times the loss at that point. His companion chart then computes EVPI from an opportunity-loss factor multiplied by the units in range and loss per unit. In other words, the calculation must use the loss relative to the break-even threshold, not the raw midpoint of the sales segment.

A concrete check shows the problem. If one slice were centered at 150,000 units, the opportunity loss for that slice would be the missed 50,000 units needed to break even, times $25 per unit = $1.25 million, before weighting by that slice’s probability. It is not simply 150,000 × probability.

So the post’s formula confuses the expected value of the underlying variable with expected opportunity loss.