Structured reasoning

Structured Reasoning

Purpose

Hard questions answered with unstructured intuition produce confident mush — conclusions that feel reasoned but can't show their work. A named framework makes reasoning inspectable: the user can see the criteria, weights, and evidence, and disagree with a specific step instead of the vibe. The skill has two halves: choosing the right frame (most of the value) and executing it honestly.

When to use this skill

• Choosing between options: tools, architectures, strategies, candidates, plans.
• Evaluating an idea, proposal, or argument — especially one you're inclined to agree with.
• Prioritization and ordering questions; recurring problems that keep coming back.
• When your own reasoning is circling: restating the same considerations without converging.

When NOT to use this skill

• Questions with verifiable answers — go verify (live-state-truth), don't deliberate.
• Low-stakes choices where any option is fine: pick one and say why in a clause.
• Don't stack frameworks for show. One frame, well executed, beats three decorative ones. Two is the practical maximum (e.g., evidence grading feeding a decision matrix).

Operating procedure

Step 1 — Classify the question, pick the frame:

Framework	Use when	Output looks like
First principles	Inherited assumptions are suspect; novel problem; "we've always done it this way"	The problem rebuilt from known constraints upward; assumptions confirmed or discarded
Tradeoff analysis	2–3 viable options differing along different dimensions	Per-option: what you gain, what you pay, when it wins
Decision matrix	3+ options × 3+ criteria; transparency matters; a group must align	Weighted criteria table with scores and the reasoning behind each score
Root-cause analysis (5 whys / causal chain)	A problem recurs despite fixes; symptom vs. disease unclear	Causal chain from symptom to actionable cause
Steelman / weakman check	Evaluating an argument or proposal; checking your own position	Strongest version of the opposing view, engaged honestly
Evidence grading	Sources conflict; claims vary in reliability	Claims ranked: verified > strong source > weak source > anecdote > speculation
Risk/reward	Bets under uncertainty; asymmetric outcomes	Upside × probability vs. downside × probability, plus reversibility
Sequencing	Order matters; dependencies and unlock effects	Ordered plan with the "why this first" argument

Step 2 — Name the frame in your output. "Treating this as a tradeoff between X and Y" — this lets the user reject the frame, which is cheaper than rejecting the conclusion.

Step 3 — Execute honestly. The failure mode of every framework is decoration: building the matrix after deciding, steelmanning weakly, grading your preferred source generously. Fill in the structure before forming the conclusion, and let the structure be allowed to surprise you.

Step 4 — Conclude with the frame's output, plus what would change it. "Option B wins on these weights; it flips to A if latency matters more than cost" — the sensitivity note is what makes the analysis reusable when circumstances shift.

Framework selection heuristics

• Stuck in circles → your frame is wrong, not your effort. Step back to Step 1.
• Everyone agrees too fast → run a steelman of the rejected option.
• The decision keeps re-litigating → decision matrix; the disagreement is usually about weights, and the matrix surfaces that.
• The plan feels overwhelming → sequencing; the question isn't "how do all of this" but "what first."
• The argument sounds great but you can't say why → evidence grading; charisma and support are different axes.

Quality bar

• The frame fits the question (a decision matrix for a two-option tradeoff is overhead; a pro/con list for a 5×5 decision is mush).
• The structure was filled before the conclusion was written — the analysis could in principle have come out the other way.
• Scores, weights, and grades come with their reasoning; a bare number is an opinion in a costume.
• The conclusion includes its sensitivity: what change of input flips it.

Common failure modes

• Framework theater: the matrix that ratifies a pre-made decision. Detectable when every criterion happens to favor the same option — real decisions have losers on some rows.
• Weakmanning: "some might say X, but that's obviously wrong" — engaging the weakest opposing argument. The steelman row exists to force the strongest.
• False precision: weights to two decimal places on gut-feel scores. Coarse honest scores (1–3, L/M/H) beat precise invented ones.
• Ignoring reversibility: treating a reversible choice with Critical-grade deliberation, or an irreversible one casually. Reversibility is a first-class input to risk/reward.
• Frame-shopping: trying frames until one endorses your prior. If two frames disagree, that disagreement is the finding to report.

Example

"Should we build the integration in-house or buy the vendor solution?"

Frame: tradeoff analysis + risk/reward (two frames, each earning its place). Build: gain control and no per-seat fees; pay 6–8 weeks of the team's roadmap and permanent maintenance. Buy: gain 2-week deployment and vendor support; pay $30k/yr and dependency on vendor roadmap. Risk asymmetry: buy is reversible (annual contract, standard data export — verified in their docs); build is not (sunk engineering). Recommendation: buy now; revisit at renewal with real usage data. Flips if: usage grows past ~$80k/yr equivalent, or the vendor can't meet the compliance requirement (unverified — check before signing).

Works with sibling skills

• verification-discipline grades the inputs; a rigorous frame over ungraded claims is precision built on sand.
• adversarial-verify attacks the conclusion afterward; the steelman row is its in-frame cousin.
• output-structuring renders the result (a decision memo, usually); failure-mode-awareness feeds the downside rows of risk/reward.
• deep-decomposition uses sequencing as its ordering engine.

Provenance and maintenance

Written 2026-07 as part of a portable, project-agnostic quality pack. Frameworks listed are standard, public reasoning tools. Re-verify by reviewing recent analyses: if conclusions never surprise (structure always ratifies the first instinct), step 3 discipline is failing. Add domain frames your work uses often (e.g., expected-value trees, pre-registered success metrics) as new rows.