Concept
Decisions
How to sort judgment before choosing an implementation.
Cover for Decisions14 concepts in this section.
All sections
Illustration for Open, closed, and closingTwo questions that sort work
12 minOpen, closed, and closing
Before you pick a model or write a prompt, two sorting questions tell you whether a decision belongs in code, in cognition, or with a person. Most architecture failures start with getting that sort wrong.
Illustration for Closed decisions belong in codeWhen rules can close the call
12 minClosed decisions belong in code
When a decision is exact, verifiable, and stable, closing it into deterministic code with tests is progress. Leaving closed work open burns money and erodes trust.
Illustration for Hard rules and soft constraintsExact rules and soft limits
12 minHard rules and soft constraints
A confirmed allergy and a general dislike of mushrooms are not the same kind of requirement. Naming which is which decides where each one is enforced and what happens when they are violated.
Illustration for Most decisions are several decisionsOne blob hides several decisions
12 minMost decisions are several decisions
Sorting one decision into open or closed is straightforward. The skill worth having is noticing that what arrived as one request is actually several decisions, with different owners and different correct homes.
Illustration for Mistakes both waysOpen where closed, closed where open
12 minMistakes both ways
Teams fear both over-automation and under-automation. Leaving closed work open wastes money and creates faux-judgment. Closing open work creates brittle harm. Good design pushes each decision toward its correct side.
Illustration for Judgment terrainSix factors around a call
14 minJudgment terrain
Fuzzy is not a terrain reading. Six factors describe the ground a decision sits on and who can safely lead it, pattern density, knowledge coverage, novelty, consequence, reversibility, and evaluability.
Illustration for Three terrain regionsMixed is where work sits
10 minThree terrain regions
Established, mixed, and frontier are three recurring combinations of terrain readings. The poles are easy to police. Mixed is where nearly all real work sits, and reading a link as mixed is only useful when it makes somebody split it.
Illustration for Judgment chainsLinked judgments need owners
10 minJudgment chains
Outcomes are produced by sequences of interpretations, checks, comparisons, and actions, not by one feature or one prompt. Mapping the chain makes hidden cognitive burden visible and gives every failure an address.
Illustration for Who should lead each judgment?Who leads which call
11 minWho should lead each judgment?
Leadership is assigned link by link, not as vague human oversight. The question is who owns the consequence of each open decision, given the terrain that link actually sits in.
Illustration for The Collaboration ContractShared rules for joint work
13 minThe Collaboration Contract
Who leads each judgment is an intention until the product specifies initiative, approval, challenge, correction, visibility, and control, link by link, not as vague human oversight.
Illustration for Human judgment that should remainCalls that stay with people
12 minHuman judgment that should remain
Some decisions stay with people even in familiar terrain, because consent, consequence, and legitimate authority have not been granted, not because models lack capability.
Illustration for Contested decisionsWhen experts disagree by design
9 minContested decisions
When competent people disagree and there is no fact of the matter, the organisation has not settled its values. Architecture must surface the disagreement, not pick silently and call it help.
Illustration for How decisions closePaths from open to closed
15 minHow decisions close
Decisions move from open to closed as cases accumulate, answers stabilise, and rules become writable. Closing is a lifecycle, not a moment of prompt writing, and cache hit rate tells you where you are.
Illustration for Choosing a supplier does not move the blameBuying does not move blame
9 minChoosing a supplier does not move the blame
Choosing among options is legitimate until the set contains paths that differ in authority, risk, or contested values. Then the selector made the higher decision, and nobody can see who owns it.