Group decision hygiene: small accuracy-committed groups

One-line summary: You cannot debias yourself; recruit two other people who have contracted to accuracy. Duke's minimum is Tetlock's three (two to disagree, one to referee), instantiated with Mertonian CUDOS norms so the group explores rather than confirms.

What it is

Duke's premise: other people spot your self-serving stories immediately, and you spot theirs. Solo "wanna bet?" is weak; a contracted group is the accountability mechanism. Size: "you actually only need three. To borrow from Phil Tetlock... you need two to disagree and one to referee."

She maps Robert Merton's CUDOS norms onto that group (she says "Kudos"):

• Communism (data-sharing): share the details you don't want to share; use a template of facts that must always be on the table so spinning is harder.
• Universalism: evaluate the claim independent of the messenger. Don't say who holds the view; or run the Obama/Trump counterfactual on the same words.
• Disinterestedness: two contaminants — the speaker's prior, and an already-known outcome (resulting). When vetting, "don't say what your belief about what the correct answer is. Just leave it out."
• Organized skepticism: ask why it isn't true. Red teams invert the "team player" problem: on a red team, disagreeing is being a team player. Anonymous dissent channels are another instantiation.

A separate warning sits upstream of the group: a leader who states a strategy with certainty "infects" the room via Gilbert's hear-and-lodge pathway. The group then confirms the boss's conclusion. Probabilistic expression is the vaccine; silence about the preferred answer is the stronger form.

Why it matters to psychology

This is applied social psychology of judgment: Mercier & Sperber-adjacent (argument is for other people), Tetlock's small forecasting teams, and Merton's scientific ethos reused as a meeting design. It is also the practical answer to Kahan's result on thinking-in-bets — if intelligence amplifies motivated reasoning, you need an external referee, not a smarter solo reasoner.

Evidence

• annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "you don't need very many people. You actually only need three. To borrow from Phil Tetlock, author of Superforecasters, you need three because you need two to disagree and one to referee."
• annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "just by expressing how you think things are going to turn out, you've now caused a virus. You've infected everybody with your belief, and they're going to tend to reason toward your conclusion because now they're going to be sort of like invited into the confirmation process with you as opposed to the exploration process with you."
• annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "When you're trying to vet a decision, don't say what your belief about what the correct answer is. Just leave it out."
• annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "whenever you feel like you don't want to share a detail, those are the details you should share."
• annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "the game for the red team is to disagree. So you've actually now to be a good team player on the red team, it's all about disagreeing."
• gary-klein in 2026-08-17-podcast-capital-allocators-best-of-pre-mortem-analysis-gary-klein (podcast / speaker claim — a different ritual than Duke’s red team, same leader-sets-the-room problem): "it's really important for the leader to come up with a real problem that hadn't been discussed before."
• paul-sonkin in 2026-08-17-podcast-capital-allocators-best-of-pre-mortem-analysis-gary-klein: "the most important factor for me is psychological safety."

Implications

• Three named people and a written contract (accuracy, accountability, open-mindedness) is the unit, not a large IC.
• Templates > exhortations: required fields beat "be transparent."
• Red-team / dissent-channel design is how you solve the team-player inhibition, not extra intelligence.

Open questions

• Does CUDOS-as-meeting-ritual actually move calibration, or is it a scientific-community metaphor that feels rigorous?
• How do you keep the third person a referee rather than a third advocate once status enters the room (boss, IC chair)?

Related

• thinking-in-bets
• resulting
• annie-duke
• investment-mistake-taxonomy
• pre-mortem-analysis — crystal-ball failure + timed writing; not a red team
• gary-klein
• paul-sonkin