research
The Same Screening Run Disagreed on Twenty-Nine Eligible Records
Nominally identical model runs agreed 91.7 percent overall but split on 94 records in a preregistered evidence-review workflow.
Summary
Nominally identical model runs agreed 91.7 percent overall but split on 94 records in a preregistered evidence-review workflow.
Twenty-nine verified eligible records appeared in only one of the paired runs. No human or model workflow recovered every eligible record, reinforcing the authors' case for validated, auditable and human-supervised use when false negatives remove evidence.
Why it matters
Nominally identical model runs agreed 91.7 percent overall but split on 94 records in a preregistered evidence-review workflow.
Limits and context
- Twenty-nine verified eligible records appeared in only one of the paired runs.
Key claims
Nominally identical model runs agreed 91.7 percent overall but split on 94 records in a preregistered evidence-review workflow.
Qualification: Twenty-nine verified eligible records appeared in only one of the paired runs.
Evidence: source-2026-08-28-020
Sources
- arXiv preprint 2608.26885arXiv · primary research
Corrections
No corrections have been recorded for this story.