Each vote carries a provenance worth tracing. In the clean baseline, each member weighs private evidence against what the others are doing, and does it optimally [5]. With no copying, each vote is a separate reading of the evidence, so counting votes amounts to counting evidence and agreement tracks accuracy all the way up [8]. Add conformity, which is the tendency to go with the majority even when your own information points the other way [2], and part of each vote is an echo of the votes before it. Unanimity is the observation that copying produces most readily, which is how the Toulouse-CNRS and Leeds team reaches a group that looks more confident and is less trustworthy [14].
The authors tested the claim across a grid of models rather than settling for one. Conformity can be written down in many ways, so they used four formulations, two information transfer schemes crossed with two mathematical models [6], then varied decision difficulty through the quality of each individual's private information, arriving at 16 cases [7]. Sixteen over four leaves four difficulty settings [15]. An effect that appears under one parameterisation of conformity is a property of that parameterisation; getting it across the grid is what makes it a property of conformity. Their stated boundary matters more to anyone reading a sign-off, because the pattern is possible except when decisions are overly simple [11]. It attaches to hard questions, which are exactly the ones where a panel's agreement is doing the most work for you.
The result fixes a direction, not a size. What is reported is a direction, plus the probability that accuracy declined as agreement increased [9], not how many points of accuracy a unanimous panel gives back relative to a nine-to-one one. It all happens inside an idealised mathematical framework whose agents are not radiologists or grant reviewers [1][5]. Harder still for practice: one observed unanimity is consistent with both regimes, the conformity-free one where it is good news [8] and the conformist one where it is not [14], leaving the two panels indistinguishable from the agreement figure alone [16].
What survives for use is narrow. Boussard's own emphasis is that agreement is a highly accessible signal in real decisions, and that a purely rational reading would wrongly treat it as an indicator of trustworthiness [13]. On this evidence, the framework changes how you should read a unanimous verdict on a hard call. How much to discount that verdict is left open.
