This appendix collects the four most-used templates. The templates themselves return to a reference-manual voice and don't extend the snark from the main text.
How to use them is simple: pick the scenario, copy the template, fill in the blanks. Don't treat templates as automatic thinking machines. Templates handle structure only; judgment is still yours.
A.1 Template A: Stable Output (six-layer framework fill-in version)
When to use: organize, rewrite, compare, or analyze tasks of medium or higher complexity.
Related chapters: Chapter 12, Chapter 13.
[Task] Write it clearly in one sentence: verb + object + scope. [Background and Materials] 1. What this material is: 2. Original material: 3. Which part to use this round: [Output Format] Language: Reader: Length: Structure: Style: Citation style: Forbidden items: [Judgment Criteria] Priority order: How to handle uncertainty: flag / fill in / skip How to handle multiple interpretations: list all / pick one / ask me [Examples and Anti-examples] Positive example: Anti-example: [Verification] Please list: - Sentences taken directly from the material - Inference sentences and the source they trace back to - The points you're least sure about - Points with no source support — to delete or flag
Common pitfalls
- Filling only the first three layers and forgetting judgment criteria and verification.
- "Format" just says "pretty," with no fields, length, or reader.
- Verification just says "please check," with no concrete items.
A.2 Template B: Few-Shot Style Control
When to use: brand voice, fixed fields, keeping style consistent across multiple pieces.
Related chapters: Chapter 9, Chapter 13.
[Task] In the style of the examples below, write ＿＿＿ pieces of ＿＿＿. [Standard Example] ＿＿＿ [Variation Example] ＿＿＿ [Boundary Example] ＿＿＿ [Anti-example] - Element: - Problem: - Contrast: - Suggestion: [New Task Topic] ＿＿＿
Common pitfalls
- Give only one example and I'll just copy it.
- Too many examples and I start overfitting on details.
- An anti-example with no "Problem / Contrast / Suggestion" tells me you didn't like something but not where to go.
A.3 Template C: Two-Stage Method
When to use: long material, complex material, tasks prone to factual drift or style drift.
Related chapters: Chapter 11, Chapter 13.
Stage 1: organize
[Task] Organize the following material into structured notes. The goal is completeness, not polish. [Material] ＿＿＿ [Format] - Core concepts: - Secondary concepts: - Context gaps: - Needs further verification: [Judgment Criteria] Completeness over polish. Don't omit source details for the sake of fluency. [Verification] For each item, mark the corresponding source paragraph or key sentence.
Mid-process human hand-off
You review the organized notes — correct errors, fill gaps, cut over-inferences.
Stage 2: write
[Task] Based on the reviewed notes below, write ＿＿＿. Only write — no more analysis. [Material] ＿＿＿ [Format] Language / Reader / Length / Structure / Style / Forbidden items [Judgment Criteria] - Stay inside the notes — nothing outside them - When a point can't be supported, leave it out [Verification] List the word count per paragraph and identify the three sentences that look most like fabrication.
Common pitfalls
- Asking for style in Stage 1 leaves the organization incomplete.
- Asking for analysis again in Stage 2 mixes the tasks back together.
- Skip the mid-process human hand-off and the two stages degrade into one slightly longer stage.
A.4 Template D: Structured Reasoning
When to use: analysis, comparison, inference, and judgment tasks.
Related chapters: Chapter 13.
[Task] Analyze / compare / infer ＿＿＿. [Material] ＿＿＿ [Steps] Execute in order: 1. List the bases for judgment first 2. Evaluate each dimension separately 3. Then give the conclusion 4. Finally, mark confidence and the factors that could overturn the conclusion [Judgment Criteria] - Every inference must trace back to the material - Mark uncertain points as "not in the material" or "needs further verification" [Format] Split into: Bases / Evaluation / Conclusion / Confidence
Common pitfalls
- Asking only for the conclusion makes it impossible to tell whether I'm just doing Fluent Fill-In.
- Without requiring traceback to the material, inference bases come out of thin air.
- Marking confidence only as high or low, with no mention of what would overturn it.