Measure the effect of restatement scope on omission counts

Measure how counting a fact as captured when it appears anywhere in a clinical note, rather than only in its required or expected section, changes omission rates for AI-generated clinical notes.

Background

The verification panel treats an omission candidate as captured if the corresponding fact appears anywhere in the note. Other annotation schemes assess omissions relative to a required section, so a fact placed elsewhere may still count as missing from the expected location. The paper finds that 25% of a sample of refused omission candidates were refused because the fact appeared elsewhere, but emphasizes that the broader effect of this scope choice on omission counts has not been directly measured.

References

The scope choice behind this is made both ways in print and its effect on a count has not been measured.

— One note in three: a verified census of three deployed AI scribes, and the instrument that counted it  (2608.31017 - Fox et al., 31 Aug 2026) in Section 4.3, “We audited our own refusals: most are judgement calls, and a quarter of the sample turned on restatement”