The vocabulary, without the padding.

Reasoning method

A procedure applied to a decision: a question asked relentlessly and a thing it refuses to let pass. Falsification asks what would make this wrong. Base rates asks how often this works out. A method is not a persona, a job function, or a personality type. Five Peers draws five from a library of twelve.

Decision record

A structured entry written at the time of a decision: the call, the options weighed, the load-bearing assumption, the stated confidence, and a review date. Its purpose is to survive hindsight, which quietly rewrites what you believed once you know the outcome. A chat transcript is not a decision record.

Base rate

How often something of this type turns out a given way, independent of the particulars of your case. Base rate neglect, judging a case on its vivid details while ignoring the rate, is among the most replicated findings in judgement research.

Pre-mortem

Assuming a plan has already failed and explaining why, rather than asking whether it might. Prospective hindsight of this kind increased correct identification of reasons by roughly 30% (Mitchell, Russo and Pennington, 1989), and was popularised for teams by Klein.

Second-order consequences

The consequences of the consequences. The first-order effect is the one you intended; the second is usually where the cost is, and it is generally visible in advance if anyone asks "and then what?".

Dialectical inquiry & devil's advocacy

Two structured dissent techniques. Devil's advocacy assigns someone to attack the plan; dialectical inquiry develops a counter-plan and argues it. Both produced better recommendations than consensus-seeking, and lower satisfaction (Schweiger, Sandberg and Ragan, 1986).

Calibration

The match between stated confidence and actual hit rate. If the things you call 80% likely happen about 80% of the time, you are well calibrated. Most people are overconfident, and almost nobody has checked.

Actively open-minded thinking

The disposition to seek out and fairly weigh evidence against your own position. Measurable with a short validated scale, and it predicts real belief updating and forecasting accuracy. The one dimension our onboarding measures head-on.

Bias blind spot

The finding that people identify bias in others while denying it in themselves, and keep denying it when shown evidence they have just displayed it (Pronin, Lin and Ross, 2002; preregistered replication 2024). It is why no onboarding should ask you to report your own biases.

Sycophancy

The tendency of a language model to shift its answer towards what the user appears to believe. A product of training on human preference data, since people rate agreement highly. Worst on judgement questions, where there is no ground truth to anchor against.

MCP (Model Context Protocol)

An open standard for connecting AI assistants to external tools and data through a common interface. It removes per-platform integration work. It does not make a product independent of the underlying model, and claims that it does are overselling.

Cognitive diversity

The idea that groups reasoning in different ways outperform homogeneous ones on complex problems. Plausible and widely cited, but the best-known formal result drew a peer-reviewed mathematical rebuttal in 2014, and the most quoted business study is an editorially reviewed magazine article. We treat it as a working hypothesis.