Spaces:
Running
Running
fix: align reporting cues and developer slugs
Browse files- app/page.tsx +10 -10
- components/benchmark-detail.tsx +3 -22
- components/benchmark-evaluation-card.tsx +3 -24
- lib/model-data.ts +29 -16
app/page.tsx
CHANGED
|
@@ -79,11 +79,11 @@ export default function HomePage() {
|
|
| 79 |
.slice(0, 6)
|
| 80 |
}, [models])
|
| 81 |
|
| 82 |
-
const
|
| 83 |
return [...models]
|
| 84 |
.sort((a, b) => {
|
| 85 |
-
if (b.
|
| 86 |
-
return b.
|
| 87 |
}
|
| 88 |
return b.benchmarks_count - a.benchmarks_count
|
| 89 |
})
|
|
@@ -248,22 +248,22 @@ export default function HomePage() {
|
|
| 248 |
|
| 249 |
<section className="grid gap-6 xl:grid-cols-2">
|
| 250 |
<OverviewPanel
|
| 251 |
-
eyebrow="
|
| 252 |
-
title="
|
| 253 |
-
description="Models with the
|
| 254 |
href="/models"
|
| 255 |
cta="Browse models"
|
| 256 |
>
|
| 257 |
<div className="space-y-1">
|
| 258 |
-
{
|
| 259 |
<ModelOverviewRow
|
| 260 |
key={model.id}
|
| 261 |
rank={index + 1}
|
| 262 |
model={model}
|
| 263 |
-
metricLabel="
|
| 264 |
-
metricValue={
|
| 265 |
secondaryLabel={`${model.benchmarks_count} benchmarks`}
|
| 266 |
-
highlight={model.
|
| 267 |
/>
|
| 268 |
))}
|
| 269 |
</div>
|
|
|
|
| 79 |
.slice(0, 6)
|
| 80 |
}, [models])
|
| 81 |
|
| 82 |
+
const mostReportedModels = useMemo(() => {
|
| 83 |
return [...models]
|
| 84 |
.sort((a, b) => {
|
| 85 |
+
if (b.evaluator_count !== a.evaluator_count) {
|
| 86 |
+
return b.evaluator_count - a.evaluator_count
|
| 87 |
}
|
| 88 |
return b.benchmarks_count - a.benchmarks_count
|
| 89 |
})
|
|
|
|
| 248 |
|
| 249 |
<section className="grid gap-6 xl:grid-cols-2">
|
| 250 |
<OverviewPanel
|
| 251 |
+
eyebrow="Reporting"
|
| 252 |
+
title="Most reporting organizations"
|
| 253 |
+
description="Models with the broadest spread of reporting organizations in the current corpus."
|
| 254 |
href="/models"
|
| 255 |
cta="Browse models"
|
| 256 |
>
|
| 257 |
<div className="space-y-1">
|
| 258 |
+
{mostReportedModels.map((model, index) => (
|
| 259 |
<ModelOverviewRow
|
| 260 |
key={model.id}
|
| 261 |
rank={index + 1}
|
| 262 |
model={model}
|
| 263 |
+
metricLabel="Reporting orgs"
|
| 264 |
+
metricValue={model.evaluator_count.toString()}
|
| 265 |
secondaryLabel={`${model.benchmarks_count} benchmarks`}
|
| 266 |
+
highlight={model.evaluator_count > 1}
|
| 267 |
/>
|
| 268 |
))}
|
| 269 |
</div>
|
components/benchmark-detail.tsx
CHANGED
|
@@ -15,7 +15,7 @@ import { Table, TableBody, TableCell, TableHead, TableHeader, TableRow } from "@
|
|
| 15 |
import {
|
| 16 |
ExternalLink, TrendingUp, Info, Database, Settings, FileCode, Building, Calendar, User, Server,
|
| 17 |
ChevronDown, ChevronUp, BarChart3, Award, AlertTriangle,
|
| 18 |
-
Cpu, Tag, Globe, Network, Activity, MessageSquare, Clock, Hash, Layers, Search, FlaskConical, Scale,
|
| 19 |
} from "lucide-react"
|
| 20 |
import type { BenchmarkEvaluation, CategoryType, EvaluationResult } from "@/lib/benchmark-schema"
|
| 21 |
import { inferCategoryFromBenchmark } from "@/lib/benchmark-schema"
|
|
@@ -363,18 +363,6 @@ function getPolicySignalLevel(score: number) {
|
|
| 363 |
}
|
| 364 |
}
|
| 365 |
|
| 366 |
-
function getSignalTone(score: number) {
|
| 367 |
-
if (score >= 0.7) {
|
| 368 |
-
return "bg-emerald-100 text-emerald-800 dark:bg-emerald-950/50 dark:text-emerald-300"
|
| 369 |
-
}
|
| 370 |
-
|
| 371 |
-
if (score >= 0.4) {
|
| 372 |
-
return "bg-amber-100 text-amber-800 dark:bg-amber-950/50 dark:text-amber-300"
|
| 373 |
-
}
|
| 374 |
-
|
| 375 |
-
return "bg-rose-100 text-rose-800 dark:bg-rose-950/50 dark:text-rose-300"
|
| 376 |
-
}
|
| 377 |
-
|
| 378 |
function getBenchmarkSpread(group: BenchmarkGroup) {
|
| 379 |
if (group.variants.length <= 1) {
|
| 380 |
return 0
|
|
@@ -845,12 +833,6 @@ export function BenchmarkDetail({ summary }: BenchmarkDetailProps) {
|
|
| 845 |
<Badge variant="outline" className="border-border/60 bg-background/80 text-[11px] uppercase tracking-[0.18em]">
|
| 846 |
Model Metadata
|
| 847 |
</Badge>
|
| 848 |
-
{policySummary.independentlyVerified && (
|
| 849 |
-
<Badge className="bg-emerald-100 text-emerald-800 hover:bg-emerald-100 dark:bg-emerald-950/40 dark:text-emerald-200">
|
| 850 |
-
<BadgeCheck className="mr-1 h-3.5 w-3.5" />
|
| 851 |
-
Independent reporting
|
| 852 |
-
</Badge>
|
| 853 |
-
)}
|
| 854 |
{formatParamsBillions(summary.model_info.additional_details?.params_billions) && (
|
| 855 |
<Badge variant="secondary" className="font-normal">
|
| 856 |
{formatParamsBillions(summary.model_info.additional_details?.params_billions)}
|
|
@@ -1871,7 +1853,6 @@ function AggregatedBenchmarkCard({
|
|
| 1871 |
|
| 1872 |
const activeFilterCount = Object.values(selectedFilters).filter((value) => value && value !== "all").length
|
| 1873 |
const leaderNormalizedScore = filteredRows[0]?.variant.normalizedScore ?? 0
|
| 1874 |
-
const signalTone = getSignalTone(group.avgNormalizedScore)
|
| 1875 |
const spread = getBenchmarkSpread(group)
|
| 1876 |
const sourceOrganizations = new Set(group.variants.map((variant) => variant.evaluation.source_metadata.source_organization_name))
|
| 1877 |
const latestTimestamp = group.variants.reduce((latest, variant) => {
|
|
@@ -1899,8 +1880,8 @@ function AggregatedBenchmarkCard({
|
|
| 1899 |
<div className="flex flex-col gap-5 lg:flex-row lg:items-start lg:justify-between">
|
| 1900 |
<div className="min-w-0 flex-1 space-y-3">
|
| 1901 |
<div className="flex flex-wrap items-center gap-2">
|
| 1902 |
-
<Badge
|
| 1903 |
-
|
| 1904 |
</Badge>
|
| 1905 |
<Badge variant="outline" className="border-border/60 bg-background/70 text-[11px] font-normal text-muted-foreground">
|
| 1906 |
{group.scoreType}
|
|
|
|
| 15 |
import {
|
| 16 |
ExternalLink, TrendingUp, Info, Database, Settings, FileCode, Building, Calendar, User, Server,
|
| 17 |
ChevronDown, ChevronUp, BarChart3, Award, AlertTriangle,
|
| 18 |
+
Cpu, Tag, Globe, Network, Activity, MessageSquare, Clock, Hash, Layers, Search, FlaskConical, Scale, BookOpenText
|
| 19 |
} from "lucide-react"
|
| 20 |
import type { BenchmarkEvaluation, CategoryType, EvaluationResult } from "@/lib/benchmark-schema"
|
| 21 |
import { inferCategoryFromBenchmark } from "@/lib/benchmark-schema"
|
|
|
|
| 363 |
}
|
| 364 |
}
|
| 365 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 366 |
function getBenchmarkSpread(group: BenchmarkGroup) {
|
| 367 |
if (group.variants.length <= 1) {
|
| 368 |
return 0
|
|
|
|
| 833 |
<Badge variant="outline" className="border-border/60 bg-background/80 text-[11px] uppercase tracking-[0.18em]">
|
| 834 |
Model Metadata
|
| 835 |
</Badge>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 836 |
{formatParamsBillions(summary.model_info.additional_details?.params_billions) && (
|
| 837 |
<Badge variant="secondary" className="font-normal">
|
| 838 |
{formatParamsBillions(summary.model_info.additional_details?.params_billions)}
|
|
|
|
| 1853 |
|
| 1854 |
const activeFilterCount = Object.values(selectedFilters).filter((value) => value && value !== "all").length
|
| 1855 |
const leaderNormalizedScore = filteredRows[0]?.variant.normalizedScore ?? 0
|
|
|
|
| 1856 |
const spread = getBenchmarkSpread(group)
|
| 1857 |
const sourceOrganizations = new Set(group.variants.map((variant) => variant.evaluation.source_metadata.source_organization_name))
|
| 1858 |
const latestTimestamp = group.variants.reduce((latest, variant) => {
|
|
|
|
| 1880 |
<div className="flex flex-col gap-5 lg:flex-row lg:items-start lg:justify-between">
|
| 1881 |
<div className="min-w-0 flex-1 space-y-3">
|
| 1882 |
<div className="flex flex-wrap items-center gap-2">
|
| 1883 |
+
<Badge variant="outline" className="border-border/60 bg-background/70 text-[11px] font-medium text-foreground/85 shadow-none">
|
| 1884 |
+
Benchmark summary
|
| 1885 |
</Badge>
|
| 1886 |
<Badge variant="outline" className="border-border/60 bg-background/70 text-[11px] font-normal text-muted-foreground">
|
| 1887 |
{group.scoreType}
|
components/benchmark-evaluation-card.tsx
CHANGED
|
@@ -5,7 +5,6 @@ import { useAudienceMode } from "@/components/audience-mode-provider"
|
|
| 5 |
import { useRouter } from "next/navigation"
|
| 6 |
import {
|
| 7 |
Award,
|
| 8 |
-
BadgeCheck,
|
| 9 |
BookOpenText,
|
| 10 |
ExternalLink,
|
| 11 |
Eye,
|
|
@@ -139,20 +138,6 @@ function getReportingSummaryLabel(data: BenchmarkEvaluationCardData) {
|
|
| 139 |
return "Aggregated reporting view"
|
| 140 |
}
|
| 141 |
|
| 142 |
-
function getReportingMixLabel(data: BenchmarkEvaluationCardData) {
|
| 143 |
-
const thirdPartyShare = Math.round(data.independent_verification_ratio * 100)
|
| 144 |
-
|
| 145 |
-
if (thirdPartyShare <= 0) {
|
| 146 |
-
return "Self-reported evidence"
|
| 147 |
-
}
|
| 148 |
-
|
| 149 |
-
if (thirdPartyShare >= 100) {
|
| 150 |
-
return "Third-party reported"
|
| 151 |
-
}
|
| 152 |
-
|
| 153 |
-
return `${thirdPartyShare}% third-party mix`
|
| 154 |
-
}
|
| 155 |
-
|
| 156 |
export function BenchmarkEvaluationCard({ data, onDelete, delayMs = 0 }: BenchmarkEvaluationCardProps) {
|
| 157 |
const router = useRouter()
|
| 158 |
const { mode } = useAudienceMode()
|
|
@@ -250,8 +235,8 @@ export function BenchmarkEvaluationCard({ data, onDelete, delayMs = 0 }: Benchma
|
|
| 250 |
tone="bg-stone-100 text-stone-900 ring-1 ring-stone-200/80 dark:bg-stone-900/40 dark:text-stone-100 dark:ring-stone-800/70"
|
| 251 |
/>
|
| 252 |
<CompactStat
|
| 253 |
-
label="Reporting
|
| 254 |
-
value={
|
| 255 |
tone="bg-emerald-50 text-emerald-900 ring-1 ring-emerald-200/70 dark:bg-emerald-950/25 dark:text-emerald-100 dark:ring-emerald-900/50"
|
| 256 |
/>
|
| 257 |
</div>
|
|
@@ -318,17 +303,11 @@ export function BenchmarkEvaluationCard({ data, onDelete, delayMs = 0 }: Benchma
|
|
| 318 |
<div className="space-y-2">
|
| 319 |
<div className="text-sm font-semibold">Reporting summary</div>
|
| 320 |
<div className="text-sm text-muted-foreground">
|
| 321 |
-
This model has reported results from {reportingSummaryLabel.toLowerCase()} across {data.benchmarks_count} benchmark{data.benchmarks_count !== 1 ? "s" : ""}.
|
| 322 |
-
{data.independent_verification_ratio > 0 && data.independent_verification_ratio < 1
|
| 323 |
-
? ` The current record mixes self-reported and third-party benchmark results, with ${Math.round(data.independent_verification_ratio * 100)}% coming from third-party reporting.`
|
| 324 |
-
: data.independent_verification_ratio >= 1
|
| 325 |
-
? " The current record is fully backed by third-party reporting."
|
| 326 |
-
: " The current record is self-reported."}
|
| 327 |
</div>
|
| 328 |
<div className="flex flex-wrap gap-2 pt-1">
|
| 329 |
{data.evaluator_names.slice(0, 2).map((evaluator) => (
|
| 330 |
<Badge key={evaluator} variant="secondary" className="font-normal">
|
| 331 |
-
<BadgeCheck className="mr-1 h-3 w-3" />
|
| 332 |
{evaluator}
|
| 333 |
</Badge>
|
| 334 |
))}
|
|
|
|
| 5 |
import { useRouter } from "next/navigation"
|
| 6 |
import {
|
| 7 |
Award,
|
|
|
|
| 8 |
BookOpenText,
|
| 9 |
ExternalLink,
|
| 10 |
Eye,
|
|
|
|
| 138 |
return "Aggregated reporting view"
|
| 139 |
}
|
| 140 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
export function BenchmarkEvaluationCard({ data, onDelete, delayMs = 0 }: BenchmarkEvaluationCardProps) {
|
| 142 |
const router = useRouter()
|
| 143 |
const { mode } = useAudienceMode()
|
|
|
|
| 235 |
tone="bg-stone-100 text-stone-900 ring-1 ring-stone-200/80 dark:bg-stone-900/40 dark:text-stone-100 dark:ring-stone-800/70"
|
| 236 |
/>
|
| 237 |
<CompactStat
|
| 238 |
+
label="Reporting Orgs"
|
| 239 |
+
value={data.evaluator_count.toString()}
|
| 240 |
tone="bg-emerald-50 text-emerald-900 ring-1 ring-emerald-200/70 dark:bg-emerald-950/25 dark:text-emerald-100 dark:ring-emerald-900/50"
|
| 241 |
/>
|
| 242 |
</div>
|
|
|
|
| 303 |
<div className="space-y-2">
|
| 304 |
<div className="text-sm font-semibold">Reporting summary</div>
|
| 305 |
<div className="text-sm text-muted-foreground">
|
| 306 |
+
This model has reported results from {reportingSummaryLabel.toLowerCase()} across {data.benchmarks_count} benchmark{data.benchmarks_count !== 1 ? "s" : ""}. Benchmark detail pages break out reporting provenance benchmark by benchmark.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 307 |
</div>
|
| 308 |
<div className="flex flex-wrap gap-2 pt-1">
|
| 309 |
{data.evaluator_names.slice(0, 2).map((evaluator) => (
|
| 310 |
<Badge key={evaluator} variant="secondary" className="font-normal">
|
|
|
|
| 311 |
{evaluator}
|
| 312 |
</Badge>
|
| 313 |
))}
|
lib/model-data.ts
CHANGED
|
@@ -371,7 +371,32 @@ function pipelineSlugify(text: string) {
|
|
| 371 |
}
|
| 372 |
|
| 373 |
function getDeveloperRouteId(developer: string) {
|
| 374 |
-
return pipelineSlugify(developer)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 375 |
}
|
| 376 |
|
| 377 |
function getModelDetailSlugCandidates(modelId: string) {
|
|
@@ -481,9 +506,7 @@ function summarizeDeveloperModels(models: IndexedModelSummary[]): DeveloperAggre
|
|
| 481 |
}
|
| 482 |
|
| 483 |
async function readDeveloperDetail(routeId: string) {
|
| 484 |
-
const direct = await
|
| 485 |
-
path.join(getDeveloperSubdirectory(), `${routeId}.json`)
|
| 486 |
-
)
|
| 487 |
|
| 488 |
if (direct?.developer && Array.isArray(direct.models)) {
|
| 489 |
return direct
|
|
@@ -499,12 +522,7 @@ async function readDeveloperDetail(routeId: string) {
|
|
| 499 |
return null
|
| 500 |
}
|
| 501 |
|
| 502 |
-
const resolved = await
|
| 503 |
-
path.join(
|
| 504 |
-
getDeveloperSubdirectory(),
|
| 505 |
-
`${getDeveloperRouteId(matchedDeveloper.developer)}.json`
|
| 506 |
-
)
|
| 507 |
-
)
|
| 508 |
|
| 509 |
if (resolved?.developer && Array.isArray(resolved.models)) {
|
| 510 |
return resolved
|
|
@@ -641,12 +659,7 @@ export async function getDeveloperList() {
|
|
| 641 |
const details = await Promise.all(
|
| 642 |
developersIndex.map(async (entry) => ({
|
| 643 |
developer: entry.developer,
|
| 644 |
-
detail: await
|
| 645 |
-
path.join(
|
| 646 |
-
getDeveloperSubdirectory(),
|
| 647 |
-
`${getDeveloperRouteId(entry.developer)}.json`
|
| 648 |
-
)
|
| 649 |
-
),
|
| 650 |
}))
|
| 651 |
)
|
| 652 |
|
|
|
|
| 371 |
}
|
| 372 |
|
| 373 |
function getDeveloperRouteId(developer: string) {
|
| 374 |
+
return pipelineSlugify(developer.trim().toLowerCase())
|
| 375 |
+
}
|
| 376 |
+
|
| 377 |
+
function getDeveloperSlugCandidates(developerOrRouteId: string) {
|
| 378 |
+
const normalized = developerOrRouteId.trim()
|
| 379 |
+
const lowercased = normalized.toLowerCase()
|
| 380 |
+
const candidates = new Set([
|
| 381 |
+
pipelineSlugify(normalized),
|
| 382 |
+
pipelineSlugify(lowercased),
|
| 383 |
+
])
|
| 384 |
+
|
| 385 |
+
return Array.from(candidates)
|
| 386 |
+
}
|
| 387 |
+
|
| 388 |
+
async function readDeveloperDetailFile(developerOrRouteId: string) {
|
| 389 |
+
for (const slug of getDeveloperSlugCandidates(developerOrRouteId)) {
|
| 390 |
+
const detail = await readJsonFile<IndexedDeveloperDetail>(
|
| 391 |
+
path.join(getDeveloperSubdirectory(), `${slug}.json`)
|
| 392 |
+
)
|
| 393 |
+
|
| 394 |
+
if (detail?.developer && Array.isArray(detail.models)) {
|
| 395 |
+
return detail
|
| 396 |
+
}
|
| 397 |
+
}
|
| 398 |
+
|
| 399 |
+
return null
|
| 400 |
}
|
| 401 |
|
| 402 |
function getModelDetailSlugCandidates(modelId: string) {
|
|
|
|
| 506 |
}
|
| 507 |
|
| 508 |
async function readDeveloperDetail(routeId: string) {
|
| 509 |
+
const direct = await readDeveloperDetailFile(routeId)
|
|
|
|
|
|
|
| 510 |
|
| 511 |
if (direct?.developer && Array.isArray(direct.models)) {
|
| 512 |
return direct
|
|
|
|
| 522 |
return null
|
| 523 |
}
|
| 524 |
|
| 525 |
+
const resolved = await readDeveloperDetailFile(matchedDeveloper.developer)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 526 |
|
| 527 |
if (resolved?.developer && Array.isArray(resolved.models)) {
|
| 528 |
return resolved
|
|
|
|
| 659 |
const details = await Promise.all(
|
| 660 |
developersIndex.map(async (entry) => ({
|
| 661 |
developer: entry.developer,
|
| 662 |
+
detail: await readDeveloperDetailFile(entry.developer),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 663 |
}))
|
| 664 |
)
|
| 665 |
|