Ten desks. Straight reporting.
Science Explained/Brief
A preprint on arXiv describes GAVEL, a protocol that uses a large language model to compare two clinical timelines against the original case report. In tests across 126 reports, merged timelines were preferred in 77.0% of comparisons and reduced discrepancies from 7.63 to 0.85 per report.
The work is a preprint, meaning it has not yet been peer-reviewed. The authors evaluated an event matcher, reviewed 2,738 findings from two models, and ranked six LLM extractors and two human annotators. Manual review confirmed 89.4% and 88.6% of findings. The authors say GAVEL supports report-based comparison and revision without treating either timeline as ground truth.
Each line below is stated by the page named above it, and only by that page.
Stated by arXiv
This preprint offers a method for evaluating timeline extraction without a perfect reference, but its results are preliminary and have not been peer-reviewed.
What the reporting says: GAVEL is an LLM judge protocol that compares two timelines with the case report and returns a discrepancy type, verdict, and report passage for each difference.
Found through monitored feeds: arXiv. A feed address is how the desk finds a story, not the evidence for it — the sources listed above are the pages the facts were taken from.
How this story was checked. Written from the 1 page listed above, stored 15 September 2026; claims checked against that stored text on 15 September 2026.