A rule that solves one problem can damage another
Most weak writing advice takes a rule that solves one problem and applies it everywhere.
- "Write for yourself" can be good advice during private discovery and bad advice when publishing a technical explainer.
- "Cut ruthlessly" can improve a bloated first draft and damage a source-faithful reference document.
- "Show, don't tell" can improve a scene and ruin an argument that simply needs to state its conclusion.
- A short sentence can be clear, or merely abrupt.
- A nut graf can orient the reader, or interrupt a narrative that already makes its promise clear.
The operating principle of this course is therefore:
First identify the job the writing is doing. Then apply the rule that fits that job.
Five jobs that are often mixed together
The course separates five jobs:
- Discovery: What do I actually think?
- Reader design: What does this reader need, and what should change for them?
- Architecture: In what order should ideas, evidence, scenes, and explanations arrive?
- Prose: How can the reader understand each sentence with minimal needless effort?
- Revision and verification: Is the piece true, supported, properly weighted, source-faithful, and doing what I intended?
These jobs interact, but they should not be confused. A beautiful sentence cannot repair a weak claim. A strong claim can still be unreadable. A compelling story can still mislead. A correct fact can still be irrelevant to the reader.
AI can make all of these failures sound finished. This course is designed to prevent that last failure in particular.
What Version 2 added
Version 2 drew on two new sources: your highlights from The Craft of Science Writing and your study of how Howard Marks builds his investment memos. The goal was not more rules. It was to strengthen places where the first version was thin.
The main additions are these:
- Topic → occasion → live question: a subject is not yet a piece. You need a reason the inquiry exists and a question that creates pressure.
- Question-ladder structure: analytical essays can advance through the questions a serious reader would ask, instead of being divided into static categories.
- Visible logic: claims connect explicitly to mechanisms, evidence, implications, qualifications, and reader objections.
- Evidence literacy: risk framing, statistical scepticism, source incentives, effect size, subgroup claims, and uncertainty.
- Explanation as design: explain only what the reader needs, when they need it. Do not let background crush the story or argument.
- Narrative precision: choosing the right unit of story, moving from micro to macro, capturing sensory detail, and building endings that both resolve and propel.
- Close reading as practice: studying good nonfiction as deliberate apprenticeship, not passive reading.
Some tempting imports were deliberately rejected:
- Not every feature needs a scene-led lede or an explicit nut graf.
- Analogies are not evidence.
- Statistical heuristics are not mechanical verdicts.
- Howard Marks's verbal mannerisms are not a style template.
Use close reading as apprenticeship
When you read a strong essay for craft, do not stop at whether you liked it or understood it. Read it more than once, with different questions each time:
- First pass, experience: read normally. Where did attention rise, fall, or become confused?
- Second pass, architecture: mark the occasion, live question, section jobs, question ladder, turns, countercases, nut function, and ending.
- Third pass, local craft: inspect paragraph openings, evidence placement, concrete detail, quotations, analogy, sentence rhythm, and compression.
- Fourth pass, transfer: extract one technique as a principle. Do not imitate wording. Apply the principle to a piece of your own.
The test of close reading is not whether you can describe another writer's style. It is whether you can identify a move, explain why it works, and reproduce the underlying function in different material.
Practise it in Drill 26: Close-reading apprenticeship.