The Paradox of Revision A Measurement Pipeline for Studying Argument Change and Conclusion

Whether peer review improves science is usually debated in terms of whether it changes papers' conclusions. Yet most review pressure concerns not the conclusion but the argument supporting it. We separate two things this conflates — how much a paper's argumentation changes and whether its take-home conclusion moves — across 3,599 open-review revision pairs from eLife and F1000-style venues, coding each with frontier large language models as auditable measurement instruments. Conclusion movement rose with argumentation change but only weakly: most moderate or substantial rewrites of a paper's reasoning left its conclusion stable (about 77%), and movement was overwhelmingly qualification — narrowing or hedging — rather than replacement (outright change, about 1%). Across nearly the entire open-review revision record (7,802 pairs) categorical change held near 1 in 70. Peer review thus acts mainly on the justification of conclusions, not their content, so evaluating it by conclusion change undercounts what it most does.