@redrob-labs/ui v1.8.0 · Redrob Group Design System 2026

ModelGuide

Agent chat · src/components/ModelGuide/ModelGuide.tsx

The full comparison behind the ranking: what each model was asked, what it wrote, and how it scored.

Two modes. Simple gives the order and a reason. Advanced shows the four dimension scores, the weights, the arithmetic and the confidence interval - everything needed to disagree with the ranking. A ranking that cannot be disagreed with is marketing, which is why the advanced view is a real view and not a footnote.

Every pick carries one real run: the prompt, the whole output in its own scrolling region, and a marker when the sample is illustrative rather than a verbatim capture. The output box is focusable so a keyboard can scroll it.

A task with no picks says "Not ranked yet" and when it will be, rather than rendering an empty list that looks like a loading failure.

Import

import { ModelGuide } from '@redrob-labs/ui';

Rendered

The design system's own preview case, run against this package's build. The parity gate compares this rendering to the reference bundle's.

Props

PropTypeNotes
professions GuideProfession[]
limit number
locale string
currency string
rates Record<string, number>
currencyByLang Record<string, string>
source { name?: string; edition?: string; note?: string }
weights GuideWeights
mode 'simple' | 'advanced' advanced shows the scores, the weights and the arithmetic.
defaultMode 'simple' | 'advanced'
profession string Controlled, so a ModelPicker can open the guide on its own task.
task string
defaultProfession string
defaultTask string
title React.ReactNode
lede React.ReactNode
method React.ReactNode How the testing works. Opened by default in advanced mode.
modeLabel string
simpleLabel React.ReactNode
advancedLabel React.ReactNode
professionLabel string
taskLabel string
perLabel string
topLabel React.ReactNode
promptLabel React.ReactNode
outputLabel string
illustrativeLabel React.ReactNode
calcLabel React.ReactNode
noteLabel string
useLabel React.ReactNode
runLabel React.ReactNode
methodLabel React.ReactNode
emptyTitle React.ReactNode
emptyText React.ReactNode
className string
onTaskChange (taskId: string | null, professionId: string) => void
onModeChange (mode: 'simple' | 'advanced') => void
onUse ( pick: GuidePick, context: { profession: GuideProfession; task: GuideTask; effort: EffortLevel | null }, ) => void effort is the level the reader tried on the pick, or null for the ranked one.
effortLabel string "Try another effort" in the detail; effortHint is the line beside it.
effortHint string
rankedLabel string
resetEffortLabel string
effortNote (context: EffortNoteContext) => React.ReactNode
languages Array<{ value: string; label: string }> The language the work is done in, which re-ranks the picks. Shown as a third select when two or more.
language string
defaultLanguage string
onLanguageChange (language: string) => void
languageLabel string
outputs Array<{ value: string; label: string }> What the work has to produce (documents, presentations, graphics...), which re-ranks the picks. Shown as a select when given, listing "anything" and then only the outputs the current task has a ranking for. Named deliverable* because outputLabel already labels a pick's sample output.
output string Controlled output; '' is "anything".
defaultOutput string
onOutputChange (output: string) => void
deliverableLabel string
anyOutputLabel string The first option, which ranks for the task as a whole.
benchmarkLabel React.ReactNode The badge on a benchmark pick, and the line its detail shows in place of the "use" action.
benchmarkNote React.ReactNode
kindLabels Partial<Record<GuideKind, string>>
comingSoonLabel React.ReactNode
missingLabel React.ReactNode Beside a tool the harness has no equivalent for. Sits with comingSoonLabel, not hardcoded.
toolsLabel React.ReactNode
sourcesLabel React.ReactNode
rangeLabel string "Likely between" before the monthly range.
harnessLabel (harness: string) => React.ReactNode Where a pick runs, e.g. (h) => \on ${h}\`` (the default).
rankLabel (place: number, task: string) => React.ReactNode The detail's rank line, e.g. (place, task) => \#${place} for ${task}\`` (the default).
effortUnit string The word after an effort level, effort by default; '' shows the level alone.
summary 'effort' | 'glance' What each row shows under its name. effort (the default) is the effort meter. glance is the comparison strip: quality, reliability, speed and value as five-step meters against the best pick on this task, under one set of column heads, with "Tied with #1" where the 95% intervals overlap and "Best value" on the most quality per dollar. With glance the effort is written out in the detail instead of metered, because a maker's own scale says little next to another maker's.
glanceLabels GlanceLabels Column heads, accessible names and tooltips for the strip. English by default.
tiedLabel React.ReactNode
bestValueLabel React.ReactNode
thinkingLabel string Before the written-out effort in the detail with summary="glance", Thinking by default.
sampleLabels GuideSampleLabels Headings and words for a sample with a task card and scorecard. English by default.
map boolean In advanced mode, a quality-against-price chart above the list. Each dot opens its pick.
mapLabels GuideMapLabels

Exported types

GuideKind, GuideStep, GuideSource, GuideTool, GuidePick, GuideTask, GuideProfession