Skip to main content

Documentation style spec

v0.1 · The single, operational spec for how Gridshift docs are written and structured. It governs everything under docs/ (Mintlify). Voice and brand rules live in BRAND.md (repo root) — this file is the docs-specific operationalization of them, and the checklist the /docs-audit skill enforces. If a rule here and BRAND.md ever disagree, BRAND.md wins on voice; this file wins on structure and Mintlify usage.

1. Voice, in one paragraph

Docs are the literal end of register by altitude — no marketing, ever. Write to a producer who is technical and impatient with fluff. Present tense, imperative for actions (“Press ⌘U”, not “You can press ⌘U”). US English (color, behavior, optimize, canceling, gray). Bold a domain term once, on first use, then plain. Cut the first sentence if it’s throat-clearing. No exclamation marks, no emoji.

2. Page anatomy

Every feature page follows the same spine:
  1. Frontmatter (see §3).
  2. One-sentence definition — opens the page, bolds the key term once.
    Quantize snaps the start of notes and audio transients to the nearest grid position.
  3. How it works — the mental model, briefly. Why it behaves as it does.
  4. Task sections — one per thing the user does, task-oriented sentence-case headings (“Quantize a selection,” not “Quantization”). Steps inside.
  5. Shortcuts — a table, when the feature has them.
  6. Related / caveats — cross-links and edge cases, via callouts.
Concept pages (concepts.mdx, introduction.mdx) explain the model and may skip the task sections. Shortcut-reference pages are primarily tables. Headings: sentence case, task-oriented, verb-first where it’s an action. Never Title Case a heading (feature names inside prose stay Title Case — they’re proper nouns; see §6).

3. Frontmatter

reviewed is the freshness signal — it means last checked for accuracy, which is what predicts staleness. It is not an edit date: git already records edits, and Mintlify shows a git-based “last updated” (metadata.timestamp). Never hand-set reviewed — a date no audit produced is a claim no one checked.

4. Components — when to use which

Use the smallest thing that does the job. A sentence beats a callout; a callout beats a section. Never stack callouts, never decorate. <Steps> and <Warning> are sanctioned but not yet used across the docs — introduce them where they fit; don’t retrofit mechanically.

5. Conventions

  • Shortcuts: real glyphs in inline code, no +: `⌘U`, `⌥A`, `⇧⌘D`. Modifier order: ⌃ ⌥ ⇧ ⌘.
  • Command Palette paths: Command Palette → **Set Record Quantize** — bold the command name.
  • Feature names: exact canonical casing from the BRAND.md glossary. Never lowercase or hyphenate a variant. Platform names keep their casing: MIDI, iOS, iCloud, macOS, MCP.
  • Cross-links: relative, by canonical name — [Groove & Swing](/features/groove).
  • Numbers & units: numerals for measured values, space before the unit — 48 kHz, 256 frames, 5 ms.
  • Defaults: always state the current default, and mark it — **None** (default).

6. Accuracy contract

This is what makes a doc correct, not just well-formed. Every one of these is a checkable claim and must match the current code:
  • Keyboard shortcuts and modifier combos.
  • Default values and the set of available options / modes / enum cases.
  • Parameter, control, and setting names (as shown in the UI).
  • Command Palette command names and menu paths.
  • File locations, formats, and limits stated as fact.
  • Stated behavior that the code actually implements.
When a claim can’t be verified from code (subjective/behavioral prose), it is flagged for human review, not silently rewritten.

7. Definition of done

A page passes audit when all of these hold:
  • Frontmatter has title + description; description is one sentence, sentence case, US English.
  • Opens with a one-sentence definition; key term bolded once.
  • Headings are sentence case and task-oriented.
  • Procedures use <Steps>; callouts follow §4; no stacked/decorative callouts.
  • Shortcuts use glyphs in inline code; feature names match the glossary; US English throughout.
  • Every checkable claim (§6) matches current code, or is flagged for review.
  • Length fits the content — a feature page is typically ~40–100 lines. No padding.
  • Cross-links resolve and use canonical names.
Last modified on August 1, 2026