Real-Stakes Practice and Identity-Protective Cognition

Question. Does real-stakes practice protect judgment on identity-charged questions, the kind where sharper reasoning usually deepens bias instead of reducing it?

Analogy. On a question that marks whose side you are on, reasoning works like a sports pundit's: the sharper you are, the better you argue your team's call, so more skill makes the bias worse, not better. Put your own money on being right and the pundit goes quiet, because now the payout comes from reality, not loyalty. The study asks whether real-stakes practice is that wager, flipping the smartest people from their tribe's best advocates into accurate ones.

What's at stake. The intuitive defence against manipulation is to teach people to reason better. Kahan showed that on identity-charged facts this backfires: the most numerate partisans are the most polarized, because they turn their reasoning skill to defending the answer their group expects. The reason is an incentive, not a deficit. On a collective fact a person pays no personal price for being wrong and a steep social price for breaking with their group. If that is the whole story, manipulation on the questions that matter most is close to unfixable by training. Missionloops bets there is a fix Kahan did not test: change the incentive. When the problem is the operator's own and the cost of being wrong is real and lands on them, reality sets the standard rather than the tribe.

The two answers it decides between. Either identity-protective cognition is effectively fixed, so trained operators polarize on charged material as much as numerate controls and reasoning skill keeps making it worse; or real-stakes contact removes the incentive that drives it, so trained operators converge toward the accurate answer on charged questions where numerate controls diverge. The study measures polarization under each and tells them apart.

What a null result would mean. If trained operators polarize on identity-charged material as much as matched high-numeracy untrained controls, the cure's core mechanism fails. That is a finding about the theory the proposal rests on, not a sign the platform was built wrong, and it would force the real-stakes claim to be retracted rather than defended.

Why this matters to defence. This is the empirical foundation under the cognitive-warfare countermeasure: it tests whether the only defence the proposal offers actually holds on the contested, identity-loaded questions an adversary targets (DRDC Objective 6, the cognitive level of combat), and whether decision-support training should be built around real stakes rather than classroom reasoning drills for analysts and personnel facing contested information (DRDC Objective 3). A positive result is doctrine: on identity-charged matters, train the stake, not the skill.

How we would run it. Give people the same number-heavy problem twice: once with a neutral label, once labelled as a hot political issue (the Kahan paradigm). Three groups receive it: Missionloops-trained operators, untrained people matched for numeracy, and a no-platform control. We measure each person's numeracy too, because the whole test is whether training flattens the pattern Kahan found, where the sharpest people polarize the most on the charged version. A wargame on real problems under genuine stakes supplies the real-stakes layer; an agency-disposition measure, where available, rides along to rule out general ability. The no-platform control separates "Missionloops specifically" from "any practice at all."

Earliest start. Stage 5: the study needs a cohort of operators who have completed real-stakes practice.