Diagnostics, ML & riskGenerated from the public API

PrecisionRecallChart

Check whether a model or a process is sane, from values you computed.

@luc/ui/charts

What it shows

Compare precision-recall trade-offs for binary classification, especially when positive cases are rare.

Live preview

Recall on x and precision on y, with the prevalence drawn as the dashed no-skill line.

<PrecisionRecallChart
  curves={[
    {
      key: "model",
      label: "Model",
      tone: "accent",
      points: [
        { x: 0.06, y: 0.94 },
        { x: 0.2, y: 0.88 },
        { x: 0.38, y: 0.79 },
        { x: 0.58, y: 0.68 },
        { x: 0.76, y: 0.55 },
        { x: 0.9, y: 0.44 },
        { x: 1, y: 0.31 },
      ],
      averagePrecision: 0.71,
    },
  ]}
  baseline={0.18}
  xFormat={{ kind: "percent", decimals: 0 }}
  yFormat={{ kind: "percent", decimals: 0 }}
  height={320}
  ariaLabel="Precision against recall"
/>

Controls

Height

320px

A chart measures its container, so it always needs a definite height.

State

The surface the chart is in — loading, empty and error are drawn by the component.

Accessible name

A drawing is not readable by a screen reader, so it needs a name that states what it shows.

The same fixture as the live preview above; the controls change how it is drawn, not what it measures.

Code

import { PrecisionRecallChart } from "@luc/ui/charts";

<PrecisionRecallChart
  curves={[
    {
      key: "model",
      label: "Model",
      tone: "accent",
      points: [
        { x: 0.06, y: 0.94 },
        { x: 0.2, y: 0.88 },
        { x: 0.38, y: 0.79 },
        { x: 0.58, y: 0.68 },
        { x: 0.76, y: 0.55 },
        { x: 0.9, y: 0.44 },
        { x: 1, y: 0.31 },
      ],
      averagePrecision: 0.71,
    },
  ]}
  baseline={0.18}
  xFormat={{ kind: "percent", decimals: 0 }}
  yFormat={{ kind: "percent", decimals: 0 }}
  height={320}
  ariaLabel="Precision against recall"
  state="ready"
/>

Not controllable here

series / rows — The numbers are the caller's; the fixture above is the example.

Copy for an AI agentA specification an agent can act on: the import, the setup, the chosen props, the data contract and the accessibility expectation.

Implement `PrecisionRecallChart` from Luc UI. It is: Check whether a model or a process is sane, from values you computed.

Import it with: import { PrecisionRecallChart } from "@luc/ui/charts";
Requires: react ^19.0.0; react-dom ^19.0.0; react-hook-form ^7.0.0 (optional, for @luc/ui/forms/react-hook-form); tailwindcss ^4.0.0; node >= 22 (repository tools). Import the stylesheet once at the app root and wrap the tree in `LucProvider`. Both come from the package that publishes @luc.

Chosen configuration: `height={320}`, `state="ready"`, `ariaLabel="Precision against recall"`.

Accessibility: keep the component's own semantics. Do not re-implement its ARIA wiring, do not add roles it does not declare, and make sure every interactive element is reachable and operable from the keyboard with a visible focus state.
Styling: use Luc UI design tokens and the component's own props. Do not restyle with raw colour literals and do not override the component's internal structure with page CSS.
Motion: respect `prefers-reduced-motion`. The component already does, so do not disable it.

Reproduce this state:
```tsx
<PrecisionRecallChart
  curves={[
    {
      key: "model",
      label: "Model",
      tone: "accent",
      points: [
        { x: 0.06, y: 0.94 },
        { x: 0.2, y: 0.88 },
        { x: 0.38, y: 0.79 },
        { x: 0.58, y: 0.68 },
        { x: 0.76, y: 0.55 },
        { x: 0.9, y: 0.44 },
        { x: 1, y: 0.31 },
      ],
      averagePrecision: 0.71,
    },
  ]}
  baseline={0.18}
  xFormat={{ kind: "percent", decimals: 0 }}
  yFormat={{ kind: "percent", decimals: 0 }}
  height={320}
  ariaLabel="Precision against recall"
  state="ready"
/>
```

Usage

The values are prepared by the caller: this component draws the points it is given and computes no statistic of its own, so units, scaling and any summary have to be settled before the data reaches it.

It is drawn by the ECharts engine.

Worked examples with real data are on the family overview: Diagnostics, ML & risk

import { PrecisionRecallChart } from "@luc/ui/charts";

<PrecisionRecallChart
  curves={[
    {
      key: "model",
      label: "Model",
      tone: "accent",
      points: [
        { x: 0.06, y: 0.94 },
        { x: 0.2, y: 0.88 },
        { x: 0.38, y: 0.79 },
        { x: 0.58, y: 0.68 },
        { x: 0.76, y: 0.55 },
        { x: 0.9, y: 0.44 },
        { x: 1, y: 0.31 },
      ],
      averagePrecision: 0.71,
    },
  ]}
  baseline={0.18}
  xFormat={{ kind: "percent", decimals: 0 }}
  yFormat={{ kind: "percent", decimals: 0 }}
  height={320}
  ariaLabel="Precision against recall"
/>

Data contract

`curves` contain ordered recall/precision `points`; `baseline` supplies the prevalence no-skill level; x and y formats declare the probability convention.

Important configuration

Label the prevalence baseline, format both axes as percentages, and keep the curve order in the supplied threshold direction.

Guidelines

When not to use

Do not compare PR curves from datasets with different prevalence without saying so; the baseline and meaning of precision change.

Configuration

Exported symbols
Export	Kind	Signature
`PrecisionRecallChart`	14 documented props	`({ curves, baseline, xFormat, yFormat, title, description, state, height, className, ariaLabel, ariaDescription, onReady, onEvent, eager, }: PrecisionRecallChartProps) => import("@types/react").JSX.E…`