information-design

Information Design

What a surface says, in what order, at what density. The bar: succinct, organized, dense where the user's task benefits — never sparse for aesthetics, never dense for its own sake.

Use This Skill When

• Laying out a screen, dashboard, table, CLI output, or report
• Judging whether a surface is too dense or too diluted
• Tightening labels, microcopy, headings, or empty-state text
• Deciding between table, list, cards, or prose for a body of information

Example trigger phrasings: "is this too dense", "this feels cluttered", "organize this screen", "tighten this copy", "should this be a table".

Do Not Use This Skill For

• Color, spacing scales, or motion — visual-language
• Whether users can find a feature, or whether an empty state routes them to the action that fills it — discoverability (this skill owns the empty state's copy)
• Chart and data-graphic construction — /dataviz (this bar still governs density and hierarchy choices there)

Reviewable Expectations

Hierarchy

• Every surface answers "what do I look at first?" in one glance: one primary element per view, established by size/weight/position — not by adding boxes.
• Reading order matches task order: the thing the user acts on next is where their eye lands next.
• Grouping follows the user's task, not the data model or the database schema.

Density

• Calibrate to user mode: monitoring, comparison, and power-user surfaces (dashboards, tables, logs, admin panels) run dense — more rows, tighter spacing, abbreviations the audience knows. First-run, decision, and confirmation surfaces run sparse — one question at a time.
• Whitespace is a hierarchy tool, not a goal. Padding that forces scrolling or pagination on a comparison task is a defect, same as a wall of undifferentiated text on a decision task.
• Progressive disclosure: summary first, detail on demand — but never hide what the user checks every time behind a click. Frequency of access decides what is surfaced, not tidiness.
• Truncation is a last resort: wrap, reflow, or widen before ellipsizing data the user came to read. Ellipsized identifiers (IDs, paths, names) that the user must distinguish are a defect.

Copy

• Words earn their place: labels are a few words, sentences appear only where the user genuinely reads. No filler ("Please note that…", "In order to…"), no restating what the UI already shows.
• Microcopy states the action, not the category: "Save draft", not "Submit"; "Delete 3 files", not "Confirm".
• One name per concept: the same entity or action carries the same word everywhere — UI, palette entries, errors, docs. Synonym drift ("remove" / "delete" / "discard" for one operation) is a defect.
• Error text says what happened and what to do next, in that order.
• Numbers formatted for scanning: aligned decimals, thousands separators, units on the column header not every cell.

Forms and validation

• Validate on blur or submit, never per keystroke: don't red-flag a field the user is still typing in. Live-as-you-type feedback is for positive confirmation (name available, password strong enough), not for errors-in-progress.
• Errors sit inline at the offending field; any summary links to them.
• Failure never destroys input: the form re-renders with every entered value intact (interaction-speed owns state preservation).
• Ask only what can't be defaulted, remembered, or inferred (design-bar's Defaults question); every optional field justifies its presence.

Presentation

• Tables for comparison across items; lists for scanning one attribute; prose for narrative only. Never prose where a table fits.
• Icons accompany labels; icon-only is reserved for universally known glyphs or space-critical dense surfaces (with tooltips and accessible labels — see accessibility).

Review Method

1. Identify the surface's user mode (first-run / decision / monitoring / power-loop) — density findings are relative to mode, not absolute.
2. Squint test: blur the layout; the primary element and grouping should survive. If everything has equal weight, hierarchy is missing.
3. Read every string aloud; cut words until meaning would break.
4. For each hidden-by-default detail, ask: how often is it checked? Daily → surface it.

No built UI yet? Apply the same expectations to the spec's described behavior; evidence cites the spec section or screen region.

Findings cite the element and the expectation violated; severity follows the design-bar rule (friction × frequency — design-bar).