In plain words: AI-written doctor notes often leave out facts, and a second AI comparing note to recording barely spots missing ones, only added or altered ones. Listing the facts the recording establishes, then checking each, catches 36.9% of omissions where usual checks catch almost none.
Abstract · LLM Judges Verify Presence, Not Absence: Omission Blindness in AI Clinical Notes and What Recovers It
Ambient AI scribes draft clinical notes, and published audits find their dominant error is omission: information the encounter established that the note fails to record. The standard check is an LLM judge: a second model reads the note against the transcript and flags problems. We ask whether judges detect omissions. Public corpora cannot supply the answer key: their clinician reference notes and transcripts are materially discrepant. Our benchmark has 500 single-error note pairs from audited fact sheets, 298 with a named fact certainly absent and 202 added-or-altered controls. Across eight judge designs, paired discrimination (the flawed note below its clean twin, 0.5 a coin flip) reads 0.79-0.94 on added or altered content and 0.50-0.63 on omissions. On single notes, no design flags omissions reliably more often than perfect notes. Wording changes, voting and GEPA prompt optimisation move the operating point without creating usable detection. Restructuring the task recovers it: list the facts the transcript establishes, then check the note for each. Two methods reach it independently and trade off: a per-fact pipeline, and a GEPA-evolved prompt doing the same in one call. The pipeline's flags name the missing fact and its severity at 2.7% false alarms. The single call detects more (36.9% against 24.6%, p=0.002) at 6.2% false alarms and a tenth of the cost per note. A physician author validated 70 items and, where the two routes disagree, sided with the pipeline on 10 of 10 (p=0.002). A second clinician, not an author, graded the severity rubric blind and agrees to within a grade. On real vendor notes from a companion census no benchmark threshold transfers, but the re-calibrated single call detects more than the best of the eight at half its false-alarm rate. Omissions whose fact is restated elsewhere defeat both routes. We release the benchmark, prompts and judgements.
Sebastian Fox, Luke Markham, Ryan Lail, Michael Karotsieris
arXiv:2608.31016 · cs.CL, cs.AI · submitted Aug 31, 2026
abstract · pdf · html · 97 pages, 8 figures. Dataset: https://huggingface.co/datasets/ComposoAI/OmissionBench . Code: https://github.com/composo-ai/omission-bench . Companion paper: "One note in three: a verified census of three deployed AI scribes, and the instrument that counted it"
