Decision Analysis

Decision Analysis

The full pipeline for consequential choices: frame, generate options, model uncertainty, score, stress-test, decide, and record.

Why

High-stakes decisions fail at whichever stage was skipped — usually option generation or stress-testing. The pipeline guarantees every stage runs at a depth proportional to stakes, so a decision that matters gets the full sequence rather than whichever piece felt most comfortable to the decider.

Use when / Don't use when

• Use when: [[reversibility-classification]] grades the decision costly-reversible or worse, and the options are genuinely open rather than pre-determined.
• Don't use when: the door is two-way — decide fast and review later — or one option already dominates on every criterion.

Inputs → Outputs

• Inputs: a consequential, non-obvious choice.
• Outputs: a decision with documented rationale, the losing options and why they lost, sensitivity flags, tripwires, and a journal entry for later scoring.

Principles

• Process weight scales with irreversibility — this pipeline is not owed to every decision, only the ones that earn it.
• Option generation is where decisions are actually won; evaluation just confirms what generation already made possible.
• Uncertainty gets modeled explicitly, not smoothed into adjectives like "likely."
• The decision isn't done until its revisit conditions are installed — a decision with no way to be reopened silently becomes dogma.

Procedure

1. Run reversibility-classification; confirm this pipeline is warranted and set the process budget accordingly.
2. Run decision-framing: the question, decider, criteria, deadline, and default path if no decision is made.
3. Run solution-space-mapping; ensure at least three genuinely distinct options exist, plus the default.
4. Model uncertainty: for each option, identify the key uncertain outcomes with probabilities, using expected-value where payoffs are commensurable.
5. Evaluate via weighted-scoring across criteria; run its sensitivity step hard — which single judgment flips the result?
6. Stress-test the leading option: run premortem and second-order-scan. Amend the option or switch to the runner-up if the wounds found are fatal.
7. Decide. Write the rationale, name the runner-up, and state why the runner-up lost — that's what gets revisited if conditions change.
8. Install tripwires: the specific observations that would reopen the decision.
9. Log via decision-journaling with calibrated confidence from confidence-calibration, for later scoring against the outcome.

Common mistakes

• Running the full pipeline on two-way doors, turning process into procrastination.
• Skipping step 6 because the leading option "feels solid" — that feeling is precisely the reason to run it.
• Writing the rationale after the fact so it sounds inevitable, rather than during the actual reasoning.
• Installing no tripwires, so the decision silently becomes unquestionable dogma.

Examples

• A build-versus-buy decision on a core system.
• A market-entry decision under genuine uncertainty.
• Choosing a therapy path from mixed clinical evidence.
• A family's decision to relocate.

Related

• [[structured-problem-solving]] — the sibling composite for when the task isn't a choice among options.
• [[scenario-planning]] — used instead when the uncertainty is about the world, not the options themselves.
• [[negotiation]] — used instead when the options are held by a counterparty, not chosen unilaterally.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: skyf0xx
• Source: skyf0xx/better-thinking
• License: MIT
• Homepage: https://zenbin.org/p/better-thinking

Install and usage instructions live in the source repository linked above.