Graphein visuals — author a spec, drop it in <Chart>

One chart is one JSON object. You don't hand-write SVG or wire a charting library. You (1) shape your data into plain rows, (2) author a single Graphein ChartSpec — a type, a tidy data array, and an encoding that names the columns — and (3) render it with the bundled <Chart> component. Graphein owns axes, scales, ticks, color, the legend, tooltips, number/date formatting, responsive sizing, animation, and dark mode.

graphein is a framework-agnostic canvas engine (no React entry of its own), so this app ships a thin binding at src/components/Chart.tsx (+ useChart.ts). Always render through <Chart> — don't call render() from graphein directly in components.

Fast path

import { Chart } from '@/components/Chart';
import type { ChartSpec } from 'graphein';

const spec: ChartSpec = {
  type: 'bar',
  title: 'Revenue by region',
  data: [
    { region: 'North', revenue: 4200 },
    { region: 'South', revenue: 3100 },
    { region: 'East', revenue: 5300 },
    { region: 'West', revenue: 2600 },
  ],
  encoding: {
    x: { field: 'region', type: 'nominal' },
    y: { field: 'revenue', type: 'quantitative', format: '$,.0f' },
  },
};

export function RevenueCard() {
  // The chart fills its parent, so give the parent an explicit size.
  return (
    <div style={{ width: '100%', height: 360 }}>
      <Chart spec={spec} />
    </div>
  );
}

Size the parent. <Chart> fills its container (width/height: 100%); it draws nothing in a zero-height box. Wrap it in an element with a real height (fixed px, a grid/flex track, or 100% of a sized ancestor).

Keep the spec stable. Pass a memoized or module-constant spec. A brand-new object every render replays the entrance animation — use useMemo when the spec depends on props/state.

Shape data as a tidy table

Graphein expects long/tidy data: one row per observation, one column per variable. The same table drives every chart — point different channels at columns. To compare groups, add a series channel — don't pre-pivot into one column per group.

// Good (tidy): one row per month × region
[{ month: '2024-01', region: 'North', revenue: 4200 },
 { month: '2024-01', region: 'South', revenue: 3100 }]
// encoding: x=month, y=revenue, series=region  → multi-series line/bar

Rows are plain JSON — no functions, no DOM nodes, no Date objects. For time axes, pass ISO strings ("2024-01", "2024-01-15") or epoch ms and set type: "temporal".

Encoding & field definitions

Cartesian charts (line/area/bar/scatter) plus pie/heatmap/funnel/ treemap/waterfall/calendarHeatmap/dumbbell/slope/combo map columns onto channels via encoding:

FieldDef = { field, type?, aggregate?, title?, format?, scale? }:

• field (required) — column name; dotted paths (a.b) read nested values.
• type — quantitative | temporal | ordinal | nominal (inferred when omitted; set it explicitly for time axes and to avoid surprises).
• format — a format hint for labels / ticks / tooltips.
• aggregate — sum | mean | avg | min | max | count | countDistinct | median | first | last when grouping.

gauge and bullet use a top-level value: { field } instead of encoding.

Chart-type catalog

Pick the closest type; there is no custom-chart escape hatch — re-express exotic shapes with the nearest supported type.

• line / area — x, y, optional series. line: points?, area?, curve?. area: stack? (totals). curve: linear|monotone|step|stepBefore| stepAfter|catmullRom.
• bar — x, y, optional series. stack? (else side-by-side groups), cornerRadius?, orientation?: 'vertical'|'horizontal' (horizontal swaps axes; keep x=category, y=value). For top-N, sort rows by value.
• scatter — x, y, optional size (bubbles), series (colored groups).
• pie / donut — theta (value) + color (category). donut?: true | 0..1 inner-radius ratio. labels?.
• heatmap — x, y (categories) + color (measure). scheme?: blues|teal|viridis|magma|greys.
• histogram — x (the numeric measure); auto-bins. bin?: { maxbins?, step? }.
• combo (dual-axis) — shared x + layers: [{ mark:'bar'|'line'|'area', axis:'left'|'right', encoding:{ y } }] for measures on independent scales.
• funnel — stage + value. percent?: 'first'|'previous', labels?.
• treemap — category + value, optional group + color.
• waterfall — stage + value (signed). totals?: string[].
• gauge / bullet — top-level value: { field }. gauge: min?, max?. bullet: target?: { field } + encoding.label.
• calendarHeatmap — date + color. scheme?.
• slope — x (two values), y, series.
• dumbbell — category + value + group (2+ points per category).
• table — columns: TableColumn[] (field, title, format, align, width, sortable, conditionalFormat, total, …), totals?, density?.
• matrix (pivot) — rows, columns, values: [{ field, op, label?, format?, conditionalFormat?, showAs? }], subtotals?, grandTotals?.
• box / sankey / choropleth — distributions / flows (source+target+ value) / geography (geo + key + color). Rare; see the library docs.

In-spec features (declarative, on BaseSpec)

Reshape and enrich a spec without touching your data pipeline:

• transform: Transform[] — aggregate, bin, filter, fold, timeUnit, calculate run before the chart builds.
• annotations: Annotation[] — reference lines/bands/zones + point callouts, e.g. a target line: annotations: [{ type: 'line', axis: 'y', value: 5000, label: 'Target' }].
• insights: true — auto-mark the max & min (opt into outliers).
• trendline: true — linear line of best fit (line/scatter).
• facet: { field, columns? } — small-multiples trellis, one panel per category.
• title, legend, tooltip, axes, animation, description (a11y).

Value formatting

Format inside the spec (never pre-format numbers to strings). A subset of d3-format for numbers, strftime for dates:

Dates: any hint with % is a date pattern — %b %e, %Y → Jan 2, 2024 (%Y/%y year, %m month, %d/%e day, %B/%b month name, %H:%M time).

Theming

This template has no design-token system, so charts render with Graphein's built-in theme (teal accent, slate neutrals, a 10-hue categorical palette) and it is dark-mode aware. To recolor, set spec.theme (a ThemeInput) — e.g. theme: { colors: { accent: '#7c3aed' } } — or pass a palette. Keep colors in the spec; don't hand-paint marks.

Interactivity (cross-chart selections)

Charts can publish/consume named selections. Create one shared bus and pass the same store to several <Chart>s to cross-highlight / cross-filter:

import { useMemo } from 'react';
import { createSelectionStore } from 'graphein';
import { Chart } from '@/components/Chart';

const store = useMemo(() => createSelectionStore(), []);

// Publisher: clicking a bar sets the "pick" selection
<Chart store={store} spec={{ type: 'bar', data: rows,
  params: [{ name: 'pick', select: { type: 'point', fields: ['region'] } }],
  encoding: { x: { field: 'region' }, y: { field: 'revenue' } } }} />

// Consumer: dims non-matching rows when "pick" changes
<Chart store={store} spec={{ type: 'line', data: rows, highlight: { param: 'pick' },
  encoding: { x: { field: 'month', type: 'temporal' },
              y: { field: 'revenue' }, series: { field: 'region' } } }} />

filter clauses ({ param }, { field, equals|oneOf|range|contains }) subset rows. Observe changes via <Chart onSelectionChange={(name, value) => …} />.

Validate & self-check

Before shipping a spec, sanity-check it:

import { validateSpec } from 'graphein';
const { valid, errors, warnings } = validateSpec(spec);

Common failures validateSpec catches:

• Missing encoding for the type (line/area/bar/scatter need x+y; pie needs theta+color; heatmap needs x+y+color; funnel needs stage+value).
• A field that doesn't exist in the data rows (a typo silently drops the channel).
• Pre-pivoted / wide data — pass tidy rows and split with series.

Empty data renders an empty chart — show a real empty/loading state in the UI rather than shipping placeholder rows.

Gotchas

• Render through <Chart>, not render() — the binding handles mount/update/ teardown and is StrictMode-safe.
• Give the chart's parent an explicit height, or it collapses to nothing.
• Memoize specs built from props/state so they don't re-animate every render.
• No custom-chart escape hatch — re-express with the closest supported type.
• Everything is plain JSON: no functions, no DOM nodes, no Date objects.