Three volatility factors are not one factor counted three times

The review everybody gave us

We asked five separate AI systems to review the scoring model. All five came back with the same headline objection: the model triple-counts volatility. Three of the six factors — IV rank, implied against realized volatility, and return on capital — are really one factor, so collapse them into one and reweight.

Five independent reviewers agreeing is hard to dismiss. We measured it instead.

The measurement

Correlations across the candidates of one scoring version, computed on the raw inputs with no scoring curve applied — because a curve can manufacture agreement between two measurements that do not have any.

The three scored volatility factors are close to mutually orthogonal, and two of the relationships are mildly negative. Collapsing them would not remove redundancy. It would delete independent signal.

Where the objection came from

That single number, quoted without the five beside it, is what every reviewer saw. Of course a richer option pays more; that is what an option premium is. It says nothing about whether IV rank — where this name’s volatility sits in its own past year — tells you anything that the return figure has already told you. Measured, it mostly does not overlap at all.

Two things we took from it

A model can look redundant and not be. IV rank is also the single most influential factor in the final ranking, tracking the total at 0.69 within a day. Folding it into a composite would have quietly removed the thing doing most of the ordering.

Five reviewers given the same partial view are one opinion, not five. Unanimity among reviewers looking at the same incomplete number is not corroboration. When we ask anyone to review the scoring now, they get the whole matrix.