RESMP.DEV BENCHMARK INSTRUMENT / REV 01

DATA-DRIVEN SMALL MULTIPLES

Evaluation
instrument.

WHAT THIS IS

A zero-dependency chart component for publishing grouped model evaluations as consistent small multiples. Give it validated JSON; it preserves series identity, a shared scale, direct labels, print layout, and an exportable browser SVG.

01 / PURPOSE

Comparisons that stay comparable.

One series order and one scale are applied across every benchmark panel. Readers can scan model behavior across STEM, instruction following, tool use, agentic tasks, or any custom family without relearning the encoding.

02 / INPUT

Plain JSON, no chart library.

Define series once, then group named panels containing numeric values keyed by series ID. Embed the JSON in HTML or assign it through the Web Component’s config property.

03 / OUTPUT

Web, print, and SVG.

The responsive web view collapses cleanly on mobile. Print mode removes the surrounding UI. SVG export preserves the complete browser-rendered composition; see the system card for portability limits.

NO FRAMEWORKNative Web Component
NO RUNTIME DATA FETCHDeterministic embedded input
EXPLICIT MISSING DATARendered as N/A, never zero
SHARED SCALEAcross the full evaluation matrix
INPUT CONTRACT
{ title, subtitle, higherIsBetter,
  series: [{ id, label, color }],
  groups: [{ name, panels: [{
    name, values: { [seriesId]: number }
  }] }] }

Plain Web Component. No framework, external font, chart library, or build step. Drop benchmark-chart.js and benchmark-component.css beside any static report.

DOCUMENTATION

Use the instrument.