Eval results tablePro

A full experiment: a column per scorer, baseline deltas, regressions first, rows that open in place.

Markdown

On this page

ProEight cases, four scorers, changes against v12; regressions first.

Installation

$ pnpm dlx shadcn@latest add @chatcn-pro/eval-results-table

Usage

tsx

import { EvalResultsTable } from "@/components/ui/chatcn/pro/eval-results-table"

<EvalResultsTable scorers={scorers} rows={rows} baselineLabel="v12" />

Each header shows the scorer's average and its change. Rows with any scorer that moved the wrong way sort first and carry a count. Sort by any scorer from its header. Variant of the free eval-results: many scorers and a baseline.

Examples

States

ProNo baseline, and in run order.

Props

Prop	Type	Notes
`scorers`	`{ id, label, kind?, higherIsBetter? }[]`	One column each
`rows`	`EvalRow[]`	`{ id, input, output?, expected?, scores, baseline? }`
`baselineLabel`	`string`	Shown in the summary
`regressionsFirst`	`boolean`	Default true
`renderDetail`	`(row) => ReactNode`	The opened row

Accessibility

Sortable headers are buttons with aria-sort; rows open with aria-expanded.
Every change has a sign, an arrow and a hidden word.