← All methodologies

kepner-tregoe — declared by cat-harness — cited, not ingested

Applies when. A decision with several candidate options and no recurring rule — a platform choice, an architecture question, which of three fixes to take. Contextual, not default: if the criteria recur, use dmn; if the question is certainty of evidence for a recommendation, use grade.

Origin. Charles H. Kepner and Benjamin B. Tregoe, The Rational Manager (1965); The New Rational Manager (1981)

Node file: kepner-tregoe.md

Kepner-Tregoe Decision Analysis

Adopted whole, 2026-09-20. External method, rendered faithfully. Where this text and the published method differ, the published method is right and this file is wrong.

The four steps

  1. State the decision. One sentence naming the kind of thing being chosen, not the options. “Which cache strategy” — not “A or B”.
  2. Separate MUSTs from WANTs. A MUST is binary and mandatory: an option that fails one is eliminated, not penalised. A WANT is desirable and comparative.
  3. Evaluate. MUSTs first, as a filter. Then WANTs across the survivors.
  4. Assess adverse consequences of the leading option, and only then decide.

What this platform adopts, and what it does not

Adopted: the MUST/WANT split, and the adverse-consequence pass.

A MUST is a real filter and needs no arithmetic — it kills an option outright, and saying why it was killed is the record. This is the step that makes an options analysis honest: most options die on a MUST, and an analysis that scores every option on every axis hides that.

The adverse-consequence pass is the same act as opening-brief’s “what would falsify the approach”, arrived at independently. Asking it of the leader specifically is KT’s contribution: the option you are about to take is the one whose failure mode you have thought about least.

Not adopted: WANT weighting.

KT scores each WANT 1–10 for importance, scores each option 1–10 against it, and sums the products. This platform does not. The weights are invented, and the arithmetic converts a judgement into a number that reads as a measurement — a total of 7.4 against 6.9 looks like evidence and is not. That is the same failure as a count quoted from prose, which uses-editorial-review and the kg:audit reading rules already refuse.

So WANTs are stated and compared in prose, and the comparison says which way each cuts. If two options survive every MUST and the WANTs genuinely do not separate them, that is the finding — say so, and let the decision be made on something else, rather than manufacturing a tiebreak.

The shape of the output

MUSTs, the options each one eliminated, the surviving options with their WANTs argued rather than scored, the leader’s adverse consequences, and the decision. Recorded per madr when the context is a bean.

Refusals

  • No weighted totals. See above.
  • No option added to make a set look considered. A straw option is worse than two real ones: it makes the analysis look thorough while narrowing it.
  • An option eliminated on a MUST is recorded, not dropped. The next agent needs to know the dead end was entered deliberately — the same reason a bean is scrapped rather than deleted.