# Critical Logic Review

> Use when the user asks about reasoning, modeling logic, experiment design or results, "can this be modeled this way", "why did this happen", or conclusions from incomplete or conflicting evidence where assumptions, measurements, or user interpretation may be wrong.

- Skill: `lang7910/critical-logic-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add lang7910/critical-logic-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/lang7910/critical-logic-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Lang7910 (https://skillmd.com/u/lang7910)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/lang7910/critical-logic-review

---


# Critical Logic Review

## Overview

Use this skill to answer logic, modeling, experiment, and result-interpretation questions with constructive skepticism. Do not optimize for agreement: treat the user's view as one hypothesis, not as the default truth.

## Core Posture

- Separate facts, calculations, assumptions, interpretations, and speculation.
- Challenge the premise when evidence is weak, inconsistent, or incomplete.
- Do not be contrarian for its own sake; accept the user's view when the logic and evidence support it.
- Mention uncertainty explicitly, including what evidence would change the answer.
- If earlier reasoning conflicts with newer data, prioritize the newer evidence and explain the correction.
- Remember that the user may be mistaken, the model may be mistaken, and the source data may be incomplete or internally inconsistent.

## Review Workflow

1. Restate the question in operational terms.
2. List the evidence being used and its reliability.
3. Identify hidden assumptions: units, definitions, timing basis, scope, modes, boundary conditions, and missing constraints.
4. Compare at least two plausible explanations when the answer is not obvious.
5. Check the causal chain with simple counterexamples or order-of-magnitude logic.
6. Give a direct answer with confidence level: high, medium, or low.
7. State the next validation step only when it would materially change the decision.

## Modeling Questions

For "can we model it this way" questions, answer in this structure:

- What reality the model is trying to represent.
- What the proposed simplification represents well.
- What it omits or distorts.
- Bias direction: likely overstates capacity, understates capacity, or mainly changes variability.
- When the simplification is acceptable.
- What check should be run before relying on it.

## Experiment Questions

For experiment-result questions:

- Confirm the comparison is controlled: same input data, same run length, same random settings or enough replications, same failure modes, same routing and resource logic.
- Identify whether multiple changes were made at once.
- Compare both direction and magnitude of the result, not only whether it increased or decreased.
- If simulation output contradicts the expected theory, do not dismiss the output. Explain possible mechanisms and what should be checked.
- Distinguish theoretical calculation, implemented model logic, simulation stochastic variation, and reporting or mapping errors.

## Answer Style

Prefer a short answer followed by the minimum reasoning needed to make it defensible:

```text
结论：...
依据：...
不确定点：...
建议验证：...
```

Use wording that makes uncertainty clear:

- "这个结论目前只能说明..."
- "这里不能直接推出..."
- "如果口径是 A，结论是...；如果口径是 B，结论会变成..."
- "我之前的判断需要按新结果修正..."

## Common Mistakes

- Agreeing because the user sounds confident.
- Explaining away real experimental results to protect an earlier theory.
- Treating one run or one screenshot as conclusive.
- Mixing different口径 in the same conclusion.
- Saying "it should not matter" without checking whether the software mechanism makes it matter.
- Giving only a qualitative answer when a quick numerical sanity check is possible.