SQYER evaluation · built with TypeSafe Jev·FDA MAUDE via openFDA·Aug 24–30, 2026

FDA received 116,623 device reports in one week. We used Jev to ask eight questions of the narratives, as a test of the model.

Every MAUDE report carries a few coded fields and a free-text narrative. To see what TypeSafe’s Jev model can do with this kind of text, we asked it the same eight questions about each narrative: what failure the text describes, what harm it describes, where the writer places the cause, and whether the manufacturer says it could not confirm the failure. Each answer came back as a probability. This page counts the tags from that pass; it is SQYER’s evaluation of the model on public text, not a finding by SQYER or by Jev. 64,085 of the 116,623 reports had no narrative text yet in FDA’s export, so the 52,538 that did are what you see here. Click anything to filter.

Reports FDA received

116,623

logged Aug 24–30, 2026

With narrative text

52,538

45% of filings · the rest had no text in the export yet

Filed as Death (event_type)

Counting the reporter's event_type field…

Narrative describes death; checkbox does not

Comparing the death tag on each narrative with its event_type…

Loading all 52,538 reports · about 0.9 MB · every chart filters the whole set in your browser0%