ComposoAI/OmissionBench

Tasks

• text-classification

Citation text from the card

Released under CC BY 4.0. Both public corpora are published under CC BY 4.0 and are
redistributed here in derived form. Reproduce these two strings with the data:

> PriMock57, Babylon Health, licensed under CC BY 4.0, github.com/babylonhealth/primock57. Changes made: transcripts normalised; the clinician reference notes were audited against their own transcripts and repaired; single-error variants were generated from the repaired notes.

> ACI-Bench, Yim et al. 2023, licensed under CC BY 4.0 on its figshare record, doi.org/10.6084/m9.figshare.22494601 (code at github.com/wyim/aci-bench, which carries no licence file of its own). Changes made: a 48-encounter subsample drawn at seed 20260728 from the 140 encounters remaining after 67 exclusions; transcripts normalised and rendered to synthetic speech; the dictation-derived reference notes were audited against their own transcripts and repaired; single-error variants were generated from the repaired notes.

`LICENSE` is the unmodified CC BY 4.0 legal code; the attribution above is also in
`ATTRIBUTION.md`, in `transcripts/README.md`, and in the `_attribution` block at the top of
every derived JSON artefact.


Cite the judge paper for the benchmark and the judge results, the census paper for the
failure census over the three deployed scribes, and this dataset record for the data.
The code repository's Zenodo archive is citable at DOI 10.5281/zenodo.22160954, which
always resolves to the newest tagged release.

> Fox, S., Markham

All datasets marked Permissive or share-alike | JSON