What is definition sensitivity?

The same study recomputed under every defensible definition of its outcomes. An edge that only exists under one reading of the words is not an edge.

This is the exhibit that catches the most self-deception, and it is worth understanding why it exists.

"The next session takes the prior high" sounds unambiguous until you try to code it. Does it mean the price traded through the high at any point, or that it closed above it? Both are reasonable. They give different answers. If your edge is 29 points under one and 15 under the other, that is fine — it survived. If it is 29 points under one and negative under the other, you have not found an edge, you have found a definition.

The table lists every outcome in the study, and under each one every alternative definition the app can express, with the signal rate, the base rate and the gap for each. The row marked as reported above is the one the headline number came from. Under each block is the verdict: survives every definition, with the weakest reading quoted, or the failure.

That verdict is what the red reverses and grey weakens pills on the desk row are counting. Reverses means the edge points the other way under another definition. Weakens means it survives but falls below the guard floor.

Alternative definitions are not computed until asked for, because each one is a real column that has to be backfilled. The exhibit says how many are missing and offers Compute N missing definitions; the reading above is never recomputed by it, only extended.