# Case studies

> Markdown twin of [danielshanklin.com/brand/case-studies](/brand/case-studies).

A case study follows rules for honest reporting, not a visual theme. Choose the
form by the question. The house materials, disclosure standard, and page layout
stay fixed.

## The eight required parts

Every case study contains these eight moves. Their order may tighten for the
story, but none may disappear silently.

1. **Question.** Name the disputed question in one sentence. A topic is not a
   question.
2. **Answer.** Give the bounded answer before the background. State what the
   evidence does and does not establish.
3. **Scope.** Name the population, unit, geography, and observed-through date
   that bound the claim.
4. **Method.** Explain how the evidence was gathered, compared, or constructed.
   Make the method reproducible enough to challenge.
5. **Evidence.** Put receipts beside the claim they support. Numbers carry
   units, baselines, dates, and run counts.
6. **Misses.** Show failed runs, excluded records, unavailable sources, and
   material uncertainty.
7. **Counter-read.** Present the strongest reasonable alternative explanation,
   not a weak objection.
8. **Decision.** End with the action the evidence supports and the observation
   that would change the conclusion.

## Choose the form

Style follows the burden of proof. Use one primary form. If a second is
necessary, make it a clearly limited section rather than a second visual
identity.

### 01 / Claim audit

- **Question:** Is a public claim actually true?
- **Story order:** Claim → source trail → corrected measure → limits →
  citation-safe wording.
- **Proof to include:** Source list, claim-status labels, corrected chart,
  quotation guidance.
- **Live exemplar:** [AI out-scores 999 of 1,000 people](/writing/ai-out-scores-999-of-1000).

### 02 / Controlled experiment

- **Question:** Does intervention X change outcome Y?
- **Story order:** Hypothesis → baseline → protocol → results → misses →
  decision rule.
- **Proof to include:** Run counts, baseline and treatment measures,
  precommitted graders, failure ledger.
- **Published example:** [A controlled test of deleting temporary support code](/writing/delete-your-scaffolding-is-a-testable-claim). <!-- plain-english: allow-url -->

### 03 / Working proof

- **Question:** Can a concrete mechanism make an abstract problem visible or
  useful?
- **Story order:** Observed problem → built artifact → behavior → what it
  proves → use boundary.
- **Proof to include:** Live prototype, annotated still, before/after view,
  build record.
- **Live exemplar:** [Data quality is the third axis](/writing/data-quality-is-the-third-axis).

### 04 / Market reading

- **Question:** What do several fresh signals jointly establish?
- **Story order:** Units and baseline → evidence → mechanism → strongest
  alternative → falsifier → bounded decision.
- **Proof to include:** Price or contract tables, source dates, mechanism
  diagram, counter-read, watch conditions.
- **Needs an update:** [Compute is tradable](/writing/compute-is-tradable).

## Name the source status

Source coverage is separate from confidence. Label every material source with
one of these terms and an observed-through date.

- **complete (✓):** The named source was checked through the stated date and
  supports the bounded claim.
- **partial (◐):** Some required evidence was checked; the missing slice is
  named beside the claim.
- **unknown (?):** The authority surface was unavailable or too stale to
  support a claim. Unknown is not negative evidence.

## Page layout

The page should feel edited before it feels designed. Prose stays narrow;
evidence earns the wider canvas.

### Type scale

Values are font size / line height in pixels.

- Page title: **48 / 52**, **36 / 40** mobile.
- Section: **30 / 36**, serif, regular.
- Subhead: **20 / 28**, serif, regular.
- Body: **16 / 28**, sans, regular.
- Caption: **14 / 20**, sans, regular.
- Evidence label: **12 / 16**, mono, regular.
- No rendered text below 12px. A short hero datum may reach 60px; the title
  does not.

### Layout rules

- **896px:** Maximum page shell, using the Brand Center gutters.
- **66ch:** Maximum narrative measure. Charts and source ledgers may use the
  full shell.
- **1 column below 768px:** Tables become labeled rows; evidence never requires
  horizontal scrolling.
- **4 ramps:** Cream is paper, wood is ink, orange is action and data, sage is
  assent. A style never gets its own palette.
- **3 font roles:** Serif is voice, sans is work, mono is evidence. Regular and
  semibold only.
- **Motion:** Fade-up for orientation only. No scroll-jacking, kinetic
  headlines, or decorative parallax.

## Do and don't

**Don't start with a visual skin.** A “Bloomberg style” or “terminal style”
page turns decoration into borrowed authority and forces the evidence into the
costume.

**Do start with the disputed question.** Choose the form that can answer it,
then render the receipts in the same house materials every other page uses.

## Final check before publishing

- Every material claim maps to a source, method, or clearly marked conclusion.
- Every number has a unit, baseline, date, and population or run count where
  relevant.
- Misses, exclusions, unavailable evidence, and the strongest counter-read are
  visible.
- The rendered page has no text below 12px, wrong font, clipped text, or
  horizontal overflow.
- Desktop and 390px mobile screenshots have been inspected, not merely
  generated.
- Links, alt text, reduced motion, build output, and the matching Markdown have
  been verified.