A red team selected for its conviction can only return its conviction

Adversarial review is supposed to test the evidence by having someone else read it differently. But "differently" has to mean a different method, not a different prior. Staff the challengers by what they already believe and you haven't built a test, you've built a delivery mechanism. The exercise outputs the selection criteria.

The clean case is the CIA's 1976 Team B experiment. To stress-test the National Intelligence Estimates on Soviet power, the outside team was staffed entirely with hawks (Pipes, Nitze, Wolfowitz) who, in Paul Warnke's phrase, had made careers of viewing the Soviet menace with alarm. The reasoning got openly circular: Team B could find no evidence of a Soviet non-acoustic submarine-detection system, and concluded the absence of evidence implied the system was already deployed and evading detection. That is a [self-sealing hypothesis](Self-sealing hypotheses cannot be broken with more evidence) wearing a review badge. Nearly every specific Team B claim later proved wrong, always in the direction of more threat.

The polarization corrupted the home team too. Knowing B would take extreme positions, the CIA analysts dug in defensively, so the exercise didn't even measure where honest analysis sat. It measured the staffing decision, twice. The Senate review's verdict: it "did not promote dissent. To the contrary, it intimidated and stifled the expression of more balanced estimates."

Source claim: Adversarial review tests the evidence only when the rival team differs in analytical method; a team selected for its conviction can only hand that conviction back.