1. Motivation

Screening is a filter with asymmetric costs. A false accept wastes reviewer effort and author money; a false reject destroys work that had value. Single-verdict screening optimises for the average case and is least reliable exactly where the decision is contested.

2. Design

Each of 600 records was screened three times under fixed, non-overlapping remits, with no evaluator seeing another's output. We recorded verdicts, agreement, and the stated reason for each rejection, then compared against the recorded editorial outcome.

3. Findings

Unanimous rejections were almost entirely uncontested: empty, out-of-scope, or contribution-free records. Unanimous accepts were similarly safe. The informative band was split verdicts, where one remit identified a specific defect the others were not looking for — most often an evidence-base problem invisible from an abstract, or a contribution claim that did not survive restatement.

Recording dissent rather than collapsing it to a majority preserved the information that made these cases decidable.

4. Implications for editorial design

Fix the remits in advance, keep the passes independent, and publish the disagreement. A screening system that reports only a verdict throws away its most useful output.