structured-analytic-techniques

Structured Analytic Techniques

Structured analytic techniques (SATs) are step-by-step methods that externalize reasoning so it can be inspected, shared, and challenged. They come from intelligence analysis, where being wrong is expensive and the mind's shortcuts (confirmation bias, anchoring, premature closure, groupthink) are the main enemy. They apply just as well to business, product, policy, research, and personal decisions.

The core idea: don't trust a conclusion that lives only in someone's head. Lay the reasoning out on the table — the hypotheses, the evidence, the assumptions, the ways it could fail — so that flaws become visible.

How to use this skill

1. Diagnose what kind of thinking trap the user is in, then pick the matching technique from the selection guide below. Don't default to one favorite — the value of SATs comes from matching the method to the failure mode.
2. Run the technique as a structured process, doing the analytical work yourself where you can and drawing the user in where their judgment or private knowledge is needed. Externalize everything: write out the matrix, the list, the table.
3. State the conclusion together with its fragility — what would change it, how confident the analysis supports being, and what to monitor.

Most techniques here are reasoning processes you run in the conversation or capture in a document. One (Analysis of Competing Hypotheses) benefits from a scoring script; use it when the hypothesis/evidence matrix gets large.

Choosing a technique

If the situation is…	The trap is usually…	Use
Several explanations compete for the same evidence	Latching onto the first/favorite explanation	Analysis of Competing Hypotheses
A plan or forecast rests on things taken for granted	Unexamined assumptions that quietly drive the conclusion	Key Assumptions Check
A decision is made and you're about to commit	Overconfidence, planning fallacy	Premortem
A view has hardened and everyone agrees	Groupthink, confirmation bias	Devil's Advocacy / Red Team
You need to detect a change early	Missing slow-building signals	Indicators & Warnings
The future is wide open and could break several ways	False precision, single-track planning	Scenario / Quadrant Crunching (see modeling-simulation skill for quantitative scenarios)
A claim is offered as fact	Accepting assertions without testing	Quality of Information Check

When a problem has several of these features, chain the techniques: a Key Assumptions Check often feeds an ACH matrix; a Premortem often surfaces indicators worth monitoring.

Full step-by-step protocols for every technique are in references/technique-catalog.md — read it when running any technique below so you follow the established structure rather than improvising.

The four workhorse techniques

These cover the large majority of cases. Lead with them.

Analysis of Competing Hypotheses (ACH)

Use when multiple explanations or predictions are in play and the temptation is to build a case for the favored one. ACH inverts the usual habit: instead of looking for evidence that confirms a hypothesis, you look for evidence that refutes each one. The hypothesis left with the least disconfirming evidence is the most likely — and crucially, ACH exposes which single piece of evidence, if wrong, would flip the conclusion.

Process: enumerate mutually exclusive hypotheses → list all relevant evidence and arguments → in a matrix, rate how consistent each piece of evidence is with each hypothesis (Consistent / Inconsistent / Neutral) → focus on inconsistencies, because hypotheses with few inconsistencies survive → identify the most diagnostic evidence (the items that discriminate between hypotheses) → report the conclusion plus its sensitivity.

For matrices beyond ~3 hypotheses × ~6 evidence items, use scripts/ach_score.py, which tallies a weighted inconsistency score per hypothesis and flags the most diagnostic evidence. See the script's header for the input format.

Key Assumptions Check

Use before committing to any analysis, plan, or forecast. List every assumption the conclusion depends on — especially the ones nobody is saying out loud because they feel obvious. For each, ask: How confident am I this holds? What happens to the conclusion if it's false? What would tell me it's breaking? Sort assumptions into solid, uncertain, and load-bearing-but-shaky. The shaky load-bearing ones are where to focus attention, contingency planning, and monitoring.

Premortem

Use just before a decision is finalized. Imagine it's some months later and the decision failed badly. Working backward from that failure, generate concrete reasons why. The prospective-hindsight framing gives people permission to voice doubts that loyalty or optimism would otherwise suppress, and it reliably surfaces risks a "what could go wrong?" question misses. Convert the top failure modes into mitigations, owners, and tripwires.

Devil's Advocacy / Red Team

Use when a position has consensus and you want to know if it's actually robust or just unchallenged. Deliberately build the strongest possible case against the prevailing view — not a strawman, the best real argument an intelligent opponent would make. Then see what survives. Frame this explicitly as "the case others would make," so it's understood as a stress test, not your own position.

Operating principles

Externalize, always. The technique's power is in writing things down where they can be seen and argued with. A matrix or a labeled list of assumptions beats a paragraph of prose every time.

Disconfirmation over confirmation. Across nearly every technique, the move that adds value is hunting for what would prove the favored idea wrong. Lean into that discomfort; it's the whole point.

Separate evidence from inference. Keep "what we observed" distinct from "what we concluded." Many bad analyses are confident inferences resting on thin or assumed evidence. The Quality of Information Check (in the catalog) makes this explicit.

Calibrate and caveat. End with how much confidence the analysis supports and what would change it. Avoid both false certainty and uselessly mushy hedging — say what you'd bet on and what you wouldn't.

Don't let the structure become theater. If a technique isn't earning its keep for the problem at hand, switch or stop. The goal is better thinking, not completed templates.

Output

For a quick application, deliver the result inline: the matrix or assumption table, the key finding, and what would change it. For a substantial analysis the user will keep or share, offer to produce a written brief (use the docx skill) structured as: question → hypotheses/assumptions considered → analysis (with the matrix or table) → finding and confidence → what to monitor / what would change the assessment.