“Cases rose 40%.” From what? Over what period? Compared to what expectation?

A percentage without a denominator is not information, it is a shape. Forty percent of a small number is a small number. Forty percent above a seasonal norm and forty percent above last week are different claims entirely.

This is one of the few things that can be checked mechanically. Every percentage in a text either has a baseline or it does not, and the absence is visible without domain knowledge.

In a sample of articles I looked at, missing baselines were among the most frequent issues — more common than outright false claims, and much less likely to be corrected, because nothing in the sentence is false.

That asymmetry seems important: the errors that survive correction processes are the ones that are technically true.

For people who work with corrections and retractions — is there any mechanism that catches omission, or does the process structurally only handle falsity?