Update about

.gitignore

@@ -49,3 +49,11 @@ whisker-render.mjs
 output/
 public/peer-ranks.json
 **/* (1).tsx
+
+# private working notes / design specs / migration plans — not for publishing
+notes/
+docs/INTERPRETIVE_SIGNALS.md
+
+# Playwright e2e output (tests/e2e via @playwright/test)
+test-results/
+playwright-report/

README.md

@@ -16,7 +16,10 @@ tags:
 
 # Eval Cards
 
-This repository is a Next.js application for viewing and authoring AI evaluations. It provides a comprehensive platform for documenting and sharing AI system evaluations across multiple dimensions including capabilities and risks.
+A Next.js application for **viewing** AI evaluations. It is the reader frontend of the
+Eval Cards platform: it does not author or store evaluation data itself — it renders a
+materialised warehouse **view layer** produced upstream by the `eval_card_backend`
+pipeline, and deploys to the Hugging Face Space `evaleval/eval-cards` (Docker runtime).
 
 ## Project Goals
 
@@ -27,56 +30,56 @@ The Eval Cards project aims to:
 - **Support research and policy** by consolidating evaluation data in an accessible format
 - **Promote responsible AI development** through comprehensive risk assessment
 
-## For External Collaborators
+## Architecture
 
-### Making Changes to Evaluation Categories and Schema
+This app is a **read-only consumer** of a snapshot. The producer (`eval_card_backend`)
+canonicalises raw evaluation data into a typed Parquet warehouse plus a Stage J view
+layer (`*.parquet` view tables + JSON sidecars), and the frontend reads that snapshot at
+runtime via DuckDB — it performs no identity resolution or aggregation of its own. The
+view-layer column names match this app's TypeScript interfaces by contract (declared in
+`lib/view-data.ts`).
 
-All evaluation categories, form fields, and data structures are centrally managed in the `schema/` folder. **This is the primary location for making structural changes to the evaluation framework.**
+Data is selected by the `DATA_BACKEND` env var. The current path is `DATA_BACKEND=v2`,
+which reads a snapshot pointed at by `SNAPSHOT_URL` (a local `file://` path in dev, or an
+`https://huggingface.co/datasets/.../resolve/<rev>/warehouse/<snapshot_id>` URL in prod).
 
-Key schema files:
-- **`schema/evaluation-schema.json`** - Defines all evaluation categories (capabilities and risks)
-- **`schema/output-schema.json`** - Defines the complete data structure for evaluation outputs
-- **`schema/system-info-schema.json`** - Defines form field options for system information
-- **`schema/category-details.json`** - Contains detailed descriptions and criteria for each category
-- **`schema/form-hints.json`** - Provides help text and guidance for form fields
-
-### Standards and Frameworks Used
-
-The evaluation framework is based on established standards:
-- **Risk categories** are derived from **NIST AI 600-1** (AI Risk Management Framework)
-- **Capability categories** are based on the **OECD AI Classification Framework**
-
-This ensures consistency with international AI governance standards and facilitates interoperability with other evaluation systems.
-
-### Contributing Evaluation Data
+## Run locally
 
-Evaluation data files are stored in `public/benchmarks/` as JSON files. Each file represents a complete evaluation of an AI system and must conform to the schema defined in `schema/output-schema.json`.
+This repo uses **pnpm** (pinned via `packageManager: pnpm@10.25.0`).
 
-To add a new evaluation:
-1. Create a new JSON file in `public/benchmarks/`
-2. Follow the structure defined in `schema/output-schema.json`
-3. Ensure all required fields are populated
-4. Validate against the schema before submission
+```bash
+pnpm install
+```
 
-### Development Setup
+Run the dev server against a local Stage J snapshot (produced by `eval_card_backend canonicalise`):
 
-## Run locally
+```bash
+DATA_BACKEND=v2 SNAPSHOT_URL=file:///abs/path/to/warehouse/<snapshot_id> pnpm dev
+```
 
-Install dependencies and run the dev server:
+Build for production and run:
 
 ```bash
-npm ci
-npm run dev
+pnpm build          # runs scripts/cache-hf-data.mjs + scripts/build-eval-matrices.mjs, then next build
+DATA_BACKEND=v2 SNAPSHOT_URL=<file:// or HF resolve URL> pnpm start
 ```
 
-Build for production and run:
+Run the test suite (Vitest):
 
 ```bash
-npm ci
-npm run build
-NODE_ENV=production PORT=3000 npm run start
+pnpm test                  # full suite
+pnpm test -- tests/<file>.test.ts   # a single test
 ```
 
+## Configuration
+
+| Env var | Purpose |
+| --- | --- |
+| `DATA_BACKEND` | Selects the data source. `v2` (a.k.a. `stage-j`) is the current view-layer backend. |
+| `SNAPSHOT_URL` | **Required when `DATA_BACKEND=v2`** — points at a Stage J snapshot directory (`file://…` locally, or an HF `…/resolve/<rev>/warehouse/<snapshot_id>` URL in prod). |
+| `SIDECAR_CACHE_DIR` / `SIDECAR_CACHE_TTL_SECONDS` / `SIDECAR_CACHE_PURGE` / `SIDECAR_BUILD_ID` | Tuning for the JSON-sidecar fetch cache. |
+| `HF_DATA_*` (`HF_DATA_LOCAL_DIR`, `HF_DATA_OFFLINE`, `HF_DATA_CACHE_TTL_MS`, …) | Knobs for the legacy v1 Hugging Face data path; not used by the v2 backend. |
+
 ## Docker (recommended for Hugging Face Spaces)
 
 A `Dockerfile` is included for deploying this app as a dynamic service on Hugging Face Spaces (Docker runtime).
@@ -90,7 +93,8 @@ docker build -t ai-eval-dashboard .
 Run the container (expose port 3000):
 
 ```bash
-docker run -p 3000:3000 -e HF_TOKEN="$HF_TOKEN" ai-eval-dashboard
+docker run -p 3000:3000 -e HF_TOKEN="$HF_TOKEN" \
+  -e DATA_BACKEND=v2 -e SNAPSHOT_URL="<HF resolve URL>" ai-eval-dashboard
 ```
 
 Visit `http://localhost:3000` to verify.
@@ -98,7 +102,14 @@ Visit `http://localhost:3000` to verify.
 ### Deploy to Hugging Face Spaces
 
 1. Create a new Space at https://huggingface.co/new-space and choose **Docker** as the runtime.
-2. Push this repository to the Space Git (or upload files through the UI). The Space will build the Docker image using the included `Dockerfile` and serve your app on port 3000.
+2. Push this repository to the Space Git (or upload files through the UI). The Space builds the Docker image using the included `Dockerfile` and serves the app on port 3000.
 
 Notes:
-- If your build needs native dependencies (e.g. `sharp`), the Docker image may require extra apt packages; update the Dockerfile accordingly.
+- If your build needs native dependencies (e.g. `sharp`), the Docker image may require extra apt packages; update the Dockerfile accordingly.
+
+## Background: evaluation framework
+
+The evaluation categories surfaced in the cards trace to established standards — risk
+categories from **NIST AI 600-1** (AI Risk Management Framework) and capability
+categories from the **OECD AI Classification Framework** — for consistency with
+international AI governance standards and interoperability with other evaluation systems.

app/about/page.tsx

@@ -10,7 +10,7 @@ export default function AboutPage() {
 
       <main className="mx-auto w-full max-w-[64rem] px-4 pb-24 pt-12 sm:px-8">
         {/* HEADER --------------------------------------------------------- */}
-        <div className="kicker">About · Working paper v0.4</div>
+        <div className="kicker">About</div>
         <h1
           className="mt-2 mb-7"
           style={{
@@ -347,19 +347,16 @@ export default function AboutPage() {
               <div className="kicker mb-2.5">Cite as</div>
               <p className="font-mono m-0 text-[12px] leading-[1.7] text-[color:var(--fg)]">
                 EvalEval Coalition. (2026). Evaluation Cards: a reporting layer for AI evaluation
-                (Working paper v0.4, snapshot 18 Apr 2026). evalcards.evalevalai.com
+                (snapshot 5 Jun 2026). evalcards.evalevalai.com
               </p>
             </div>
             <div className="border border-[color:var(--border-soft)] p-[22px]">
               <div className="kicker mb-2.5">Submit a correction</div>
-              <p className="m-0 mb-2 text-[13.5px] leading-[1.65] text-[color:var(--fg-muted)]">
+              <p className="m-0 text-[13.5px] leading-[1.65] text-[color:var(--fg-muted)]">
                 Each record links a correction path. Disclosure gaps close as developers and
                 third parties publish; we accept patches against any (model, benchmark,
                 metric-path) tuple with a citation.
               </p>
-              <span className="font-mono text-[color:var(--fg-subtle)] text-[11px] uppercase tracking-[0.1em]">
-                corrections@evalevalai.com
-              </span>
             </div>
           </div>
         </section>

components/benchmark-detail.tsx

@@ -5,7 +5,7 @@ import Link from "next/link"
 import { usePathname, useRouter, useSearchParams } from "next/navigation"
 import { useAudienceMode } from "@/components/audience-mode-provider"
 import { EmbedButton } from "@/components/embed-button"
-import { formatDateISO, humanizeBenchmarkName, humanizeEvaluationId, routeIdToPath } from "@/lib/utils"
+import { formatDateISO, humanizeBenchmarkName, humanizeEvaluationId, routeIdFromModelId, routeIdToPath } from "@/lib/utils"
 import { Card, CardContent, CardDescription, CardHeader, CardTitle } from "@/components/ui/card"
 import { Badge } from "@/components/ui/badge"
 import { Button } from "@/components/ui/button"
@@ -779,7 +779,7 @@ function getComparisonScoreEntryForVariant(
   const modelId = row.variant.evaluation.model_info.id
   return (
     metricEntry.scores.find(
-      (score) => score.model_route_id === modelId || score.model_family_id === modelId
+      (score) => score.model_route_id === modelId || score.model_group_id === modelId
     ) ?? null
   )
 }
@@ -2733,10 +2733,19 @@ export function BenchmarkDetail({
   // Used to (a) pull our own score out of `by_model` and (b) drop ourselves
   // out of the peer score list.
   const currentModelIdentityKeys = useMemo(() => {
+    const id = summary.model_info.id || ""
+    // Include BOTH the plain id and its encoded (route_id) form: comparison
+    // score rows are keyed by the percent-encoded route, while the summary
+    // exposes the plain `id`. Without the encoded forms the current model
+    // never matches its own score row (so it isn't excluded from peers).
     const keys = new Set<string>(
       [
-        summary.model_info.id,
-        (summary as any).model_family_id,
+        id,
+        id && encodeURIComponent(id),
+        (summary as any).model_route_id,
+        (summary as any).variant_key,
+        (summary as any).model_group_id,
+        (summary as any).model_group_id && encodeURIComponent((summary as any).model_group_id),
         (summary.model_info as any).family_id,
         (summary.model_info as any).model_route_id,
         ...((summary as any).raw_model_ids ?? []),
@@ -2745,15 +2754,19 @@ export function BenchmarkDetail({
     return keys
   }, [summary])
 
-  // The primary model_route_id that keys into comparison-index.by_model for
-  // this page. Prefer an explicit route id; otherwise derive one.
+  // The model_route_id that keys into comparison-index.by_model (which is keyed
+  // by the GROUP route_id). The page renders the selected variant as `summary`,
+  // so prefer its carried-through model_route_id, then variant_key, and finally
+  // encode the plain id — never the legacy `__` form, which the v2 backend's
+  // percent-encoded keys don't use.
   const currentModelRouteId = useMemo(() => {
     const explicit =
+      (summary as any).model_route_id ||
       (summary.model_info as any).model_route_id ||
-      (summary as any).model_route_id
+      (summary as any).variant_key
     if (typeof explicit === "string" && explicit.length > 0) return explicit
     const id = summary.model_info.id || ""
-    return id.replace(/[/]/g, "__")
+    return id ? encodeURIComponent(id) : ""
   }, [summary])
 
   // Cross-suite overlaps: walk `benchmark_index[]` (already pre-filtered by
@@ -2779,7 +2792,7 @@ export function BenchmarkDetail({
       for (const row of metric.scores) {
         if (
           currentModelIdentityKeys.has(row.model_route_id) ||
-          currentModelIdentityKeys.has(row.model_family_id)
+          currentModelIdentityKeys.has(row.model_group_id)
         ) {
           if (Number.isFinite(row.score)) return row.score
         }
@@ -2985,7 +2998,7 @@ export function BenchmarkDetail({
         for (const s of metric.scores) {
           if (
             currentModelIdentityKeys.has(s.model_route_id) ||
-            currentModelIdentityKeys.has(s.model_family_id)
+            currentModelIdentityKeys.has(s.model_group_id)
           ) {
             currentRow = s
             break
@@ -3012,7 +3025,7 @@ export function BenchmarkDetail({
         const peerRows = metric.scores.filter(
           (s) =>
             !currentModelIdentityKeys.has(s.model_route_id) &&
-            !currentModelIdentityKeys.has(s.model_family_id)
+            !currentModelIdentityKeys.has(s.model_group_id)
         )
 
         const defaults = new Set<string>()
@@ -3039,11 +3052,10 @@ export function BenchmarkDetail({
           .filter((p) => selectedIds.has(p.model_route_id))
           .map((p) => ({
             modelId: p.model_route_id,
-            // Fall back to model_family_id when the registry has no display
-            // name for this model. Many score entries land here; the root cause
-            // is registry coverage. The id is
-            // already human-readable in this codebase ("anthropic/Sonnet 4.5"),
-            // so the fallback is more useful than "Unknown Model".
+            // Most comparison score rows carry an empty model_family_name, so
+            // fall back to model_family_id (always present, and already
+            // human-readable in this codebase, e.g. "anthropic/sonnet-4.5").
+            // Without this the peer bars all read "Unknown Model".
             modelName: getModelDisplayName(p.model_family_name || p.model_family_id),
             score: p.score,
             isCurrent: false,
@@ -3077,12 +3089,10 @@ export function BenchmarkDetail({
           .filter((p) => !selectedIds.has(p.model_route_id))
           .map((p) => ({
             id: p.model_route_id,
-            // Same fallback as the rendered bar (peerBars above) — when
-            // the registry lacks a display name, fall back to the
-            // model_family_id (which is already human-readable in this
-            // codebase, e.g. "anthropic/sonnet-4.5"). Without this the
-            // dropdown label resolves to "Unknown Model" even though
-            // the bar that appears after selection reads correctly.
+            // Same fallback as the rendered bar (peerBars above): score rows
+            // usually have an empty model_family_name, so fall back to
+            // model_family_id (always present, human-readable). Without this the
+            // dropdown label resolves to "Unknown Model".
             name: p.model_family_name || p.model_family_id,
             score: p.score,
             submissionCount: p.submission_count,
@@ -3924,7 +3934,7 @@ export function BenchmarkDetail({
           for (const row of siblingMetric.scores) {
             if (
               currentModelIdentityKeys.has(row.model_route_id) ||
-              currentModelIdentityKeys.has(row.model_family_id)
+              currentModelIdentityKeys.has(row.model_group_id)
             ) {
               if (Number.isFinite(row.score)) {
                 siblingScore = row.score
@@ -4395,6 +4405,8 @@ export function BenchmarkDetail({
             return (
               <div
                 key={bar.modelId}
+                data-model-bar={bar.modelId}
+                data-bar-current={bar.isCurrent ? "1" : "0"}
                 className="group flex min-w-0 flex-col items-center"
               >
                 <div className="relative flex h-44 w-full items-end">
@@ -4708,6 +4720,13 @@ export function BenchmarkDetail({
                   <span>{formatParamsBillions(summary.model_info.additional_details?.params_billions)}</span>
                 </>
               )}
+              {/* model-resolution-rework: served-by inference platform (nullable). */}
+              {summary.model_info.inference_platform && (
+                <>
+                  <span className="text-[color:var(--fg-subtle)]">·</span>
+                  <span>Served by {summary.model_info.inference_platform}</span>
+                </>
+              )}
             </div>
           </div>
 
@@ -4731,6 +4750,41 @@ export function BenchmarkDetail({
             <div className="mt-1 text-right font-mono text-[10px] uppercase tracking-[0.15em] text-[color:var(--fg-subtle)]">
               {Math.max(0, reproducibilityResultsTotal - reproducibilityGapCount)} / {reproducibilityResultsTotal} reported
             </div>
+
+            {/* model-resolution-rework: lineage + resolution provenance.
+                All fields are nullable (server-provided); the block only
+                renders rows that are present. */}
+            {(summary.lineage_origin_model_id ||
+              summary.resolution_source ||
+              summary.resolution_granularity) && (
+              <dl className="mt-4 space-y-1 text-[11px] text-[color:var(--fg-muted)]">
+                {summary.lineage_origin_model_id && (
+                  <div className="flex items-baseline justify-between gap-2">
+                    <dt className="kicker shrink-0">Base model</dt>
+                    <dd className="min-w-0 truncate text-right">
+                      <Link
+                        href={`/models/${routeIdToPath(routeIdFromModelId(summary.lineage_origin_model_id))}`}
+                        className="font-mono hover:text-[color:var(--accent)]"
+                      >
+                        {summary.lineage_origin_model_id}
+                      </Link>
+                    </dd>
+                  </div>
+                )}
+                {summary.resolution_source && (
+                  <div className="flex items-baseline justify-between gap-2">
+                    <dt className="kicker shrink-0">Resolved via</dt>
+                    <dd className="font-mono">{summary.resolution_source}</dd>
+                  </div>
+                )}
+                {summary.resolution_granularity && (
+                  <div className="flex items-baseline justify-between gap-2">
+                    <dt className="kicker shrink-0">Granularity</dt>
+                    <dd className="font-mono">{summary.resolution_granularity}</dd>
+                  </div>
+                )}
+              </dl>
+            )}
           </div>
         </div>
       </header>

components/eval-detail.tsx

@@ -33,8 +33,7 @@ import {
   DropdownMenuSeparator,
   DropdownMenuTrigger,
 } from "@/components/ui/dropdown-menu"
-import { getModelFamilyRouteId } from "@/lib/model-family"
-import { cn, formatDateISO, humanizeEvaluationId, routeIdToPath } from "@/lib/utils"
+import { cn, formatDateISO, humanizeEvaluationId, routeIdFromModelId, routeIdToPath } from "@/lib/utils"
 import {
   AlertTriangle,
   BarChart3,
@@ -1580,7 +1579,7 @@ export function EvalDetail({
                       </td>
                       <td style={{ padding: "10px 8px", color: "var(--fg)" }}>
                         <Link
-                          href={`/models/${routeIdToPath(modelResult.model_route_id ?? getModelFamilyRouteId(modelResult.model_info))}`}
+                          href={`/models/${routeIdToPath(modelResult.model_route_id ?? routeIdFromModelId(modelResult.model_group_id))}`}
                           className="hover:text-[color:var(--accent)] transition-colors"
                           style={{ color: "var(--fg)", fontWeight: 500, fontSize: 14 }}
                         >
@@ -1783,7 +1782,7 @@ export function EvalDetail({
                             )}
                             <div className="min-w-0">
                               <Link
-                                href={`/models/${routeIdToPath(modelResult.model_route_id ?? getModelFamilyRouteId(modelResult.model_info))}`}
+                                href={`/models/${routeIdToPath(modelResult.model_route_id ?? routeIdFromModelId(modelResult.model_group_id))}`}
                                 className="font-semibold text-[14px] hover:text-[color:var(--accent)] transition-colors"
                                 style={{ color: "var(--fg)" }}
                               >
@@ -2768,7 +2767,7 @@ function MultiMetricLeaderboard({
                       )}
                       <div className="min-w-0">
                         <Link
-                          href={`/models/${routeIdToPath(row.model_route_id ?? getModelFamilyRouteId(row.model_info))}`}
+                          href={`/models/${routeIdToPath(row.model_route_id ?? routeIdFromModelId(row.model_group_id))}`}
                           className="font-semibold text-[14px] hover:text-[color:var(--accent)] transition-colors"
                           style={{ color: "var(--fg)" }}
                         >

lib/backend-artifacts.ts

@@ -414,7 +414,13 @@ export interface ComparisonSubmission {
 
 export interface ComparisonScoreEntry {
   model_route_id: string
+  // The group root id, slash-form (e.g. "zhipu/glm-4-6-fc-thinking") — the field
+  // comparison-index score rows actually carry. Peer labels fall back to it when
+  // model_family_name is empty (most rows). NOTE: `model_group_id` below is NOT
+  // present on these rows; it stays declared only because other consumers read
+  // it as a `|| model_group_id` fallback (always undefined here, harmless).
   model_family_id: string
+  model_group_id: string
   model_family_name: string
   developer: string
   variant_key: string

lib/benchmark-schema.ts

@@ -261,6 +261,12 @@ export interface ModelSummaryCore extends SignalSummaries {
   total_evaluations: number
   last_updated: string
   tags_covered: EvalTag[]
+  // model-resolution-rework (additive, all nullable). Server-provided model
+  // identity provenance; surfaced on the model detail page. Carried here on
+  // the core so both the summary and variant shapes expose them.
+  lineage_origin_model_id?: string    // deepest non-variant ancestor (base model)
+  resolution_source?: string          // enum: hf | models_dev | curated | inferred | none
+  resolution_granularity?: string     // enum: variant | group | family
 }
 
 export interface ModelVariantSummary extends ModelSummaryCore {
@@ -276,7 +282,7 @@ export interface ModelVariantSummary extends ModelSummaryCore {
 }
 
 export interface ModelEvaluationSummary extends ModelSummaryCore {
-  model_family_id: string
+  model_group_id: string
   model_route_id: string
   model_family_name: string
   raw_model_ids: string[]
@@ -347,6 +353,14 @@ export interface EvaluationCardData {
   params?: string
   inference_engine?: string
   inference_platform?: string
+
+  // model-resolution-rework (additive, all nullable). Server-provided
+  // (producer view layer) — the frontend no longer computes families
+  // client-side. See notes/backend-v2-migration.md.
+  model_group_id?: string             // group canonical id (membership / grouping root)
+  lineage_origin_model_id?: string    // deepest non-variant ancestor (base model)
+  resolution_source?: string          // enum: hf | models_dev | curated | inferred | none
+  resolution_granularity?: string     // enum: variant | group | family
 }
 
 // ── Benchmark Card types (from metadata/benchmark_card_*.json) ────────────────

lib/clean-hierarchy.ts

@@ -195,7 +195,7 @@ type ComparisonIndexLike = {
         metric_name?: string | null
         scores: Array<{
           model_route_id?: string | null
-          model_family_id?: string | null
+          model_group_id?: string | null
           score?: number | null
         }>
       }>
@@ -280,7 +280,7 @@ function buildModelCoverageMap(
   for (const [evalId, entry] of Object.entries(comparisonIndex.evals ?? {})) {
     for (const metric of entry.metrics ?? []) {
       for (const row of metric.scores ?? []) {
-        const modelId = row.model_route_id || row.model_family_id
+        const modelId = row.model_route_id || row.model_group_id
         if (!modelId || row.score == null || !Number.isFinite(row.score as number)) continue
         const set = modelEvals.get(modelId) ?? new Set<string>()
         set.add(evalId)
@@ -415,7 +415,7 @@ function dedupAggregatorBenchesByScore(
       usableMetrics[0]
     const map = new Map<string, number>()
     for (const row of target.scores ?? []) {
-      const id = row.model_route_id || row.model_family_id
+      const id = row.model_route_id || row.model_group_id
       if (!id || row.score == null || !Number.isFinite(row.score)) continue
       map.set(id, row.score as number)
     }

lib/eval-processing.ts

@@ -25,6 +25,12 @@ export type { ModelEvaluationSummary }
 export interface ModelResultForBenchmark {
   model_info: ModelInfo
   model_route_id?: string
+  /**
+   * model-resolution-rework: server-provided group canonical id. Used as
+   * the routing fallback when `model_route_id` is absent — replaces the
+   * old client-side family-route computation (since removed).
+   */
+  model_group_id?: string
   score: number
   score_details: ScoreDetails
   evaluation_timestamp: string
@@ -163,6 +169,8 @@ export interface BenchmarkLeaderboardMetric {
 export interface BenchmarkLeaderboardRow {
   model_info: ModelInfo
   model_route_id?: string
+  /** model-resolution-rework: server group id, routing fallback. */
+  model_group_id?: string
   evaluation_timestamp: string
   source_metadata: SourceMetadata
   source_data: BenchmarkEvaluation["source_data"]

lib/hf-data.ts

@@ -31,7 +31,6 @@ import type {
   SourceData,
   SourceMetadata,
 } from "@/lib/benchmark-schema"
-import { getCanonicalModelIdentity, getModelFamilyRouteId } from "@/lib/model-family"
 
 // ---------------------------------------------------------------------------
 // HuggingFace dataset base URL
@@ -512,7 +511,7 @@ export interface HFModelDetail extends SignalSummaries {
     variant_label?: string
     model_route_id?: string
   }
-  model_family_id: string
+  model_group_id: string
   model_route_id: string
   model_family_name?: string
   raw_model_ids?: string[]

lib/known-developers.ts

@@ -20,12 +20,38 @@ export const KNOWN_DEVELOPER_NAMES: Record<string, string> = {
   "x-ai": "xAI",
 }
 
-// Canonical names treated as "official" model-developer organisations.
-// Seeded from KNOWN_DEVELOPER_NAMES values; extend with org names that
-// arrive already-canonical from the backend registry (no case fix needed).
+// Canonical display names treated as "official" model-developer organisations.
+// MIRRORS the registry's `canonical_orgs.kind == 'lab'` set — the `developer`
+// string the backend emits is already the org's canonical display_name, so we
+// match on it directly. Keep in sync when curated labs are added to the registry
+// (eval-card-registry/seed/orgs.yaml). Durable follow-up: have the producer emit
+// an `is_official` flag on the developer list so this hand-maintained set can be
+// retired. (KNOWN_DEVELOPER_NAMES values are a subset, included via the spread.)
 export const OFFICIAL_DEVELOPER_NAMES: Set<string> = new Set([
   ...Object.values(KNOWN_DEVELOPER_NAMES),
+  "01.AI",
+  "AI21 Labs",
   "Allen Institute for AI",
+  "BigScience",
+  "ByteDance",
+  "Databricks",
+  "EleutherAI",
+  "Google DeepMind",
+  "Hugging Face",
+  "Inception Labs",
+  "MiniMax",
+  "Moonshot AI",
+  "Nous Research",
+  "Perplexity AI",
+  "Sarvam AI",
+  "Stability AI",
+  "Stanford CRFM",
+  "Stanford University",
+  "StepFun",
+  "Technology Innovation Institute",
+  "Upstage",
+  "Writer",
+  "Z.AI",
 ])
 
 export function normalizeDeveloperName(name: string): string {

lib/model-family.ts

@@ -1,202 +0,0 @@
-import type { ModelInfo } from "@/lib/benchmark-schema"
-
-interface ParsedModelIdentity {
-  namespace: string
-  rawHandle: string
-  normalizedHandle: string
-  familySlug: string
-  familyId: string
-  familyName: string
-  variantKey: string
-  variantLabel: string
-  variantDisplayName: string
-  versionDate?: string
-  versionQualifier?: string
-}
-
-const TOKEN_CASE_MAP: Record<string, string> = {
-  ai: "AI",
-  coder: "Coder",
-  command: "Command",
-  chat: "Chat",
-  claude: "Claude",
-  gemini: "Gemini",
-  gemma: "Gemma",
-  gpt: "GPT",
-  haiku: "Haiku",
-  instruct: "Instruct",
-  instant: "Instant",
-  llama: "Llama",
-  max: "Max",
-  mini: "Mini",
-  mistral: "Mistral",
-  opus: "Opus",
-  phi: "Phi",
-  plus: "Plus",
-  preview: "Preview",
-  pro: "Pro",
-  qwen: "Qwen",
-  reasoning: "Reasoning",
-  sonnet: "Sonnet",
-  thinking: "Thinking",
-  turbo: "Turbo",
-  yi: "Yi",
-}
-
-function stripNamespace(value: string, namespace: string) {
-  const trimmed = value.trim()
-  if (!trimmed) {
-    return trimmed
-  }
-
-  const lowerTrimmed = trimmed.toLowerCase()
-  const namespacePrefix = `${namespace.toLowerCase()}/`
-
-  if (lowerTrimmed.startsWith(namespacePrefix)) {
-    return trimmed.slice(namespacePrefix.length)
-  }
-
-  return trimmed
-}
-
-function getNamespace(modelInfo: ModelInfo) {
-  const idParts = modelInfo.id.split("/")
-  if (idParts.length > 1 && idParts[0]) {
-    return idParts[0].trim().toLowerCase()
-  }
-
-  return (modelInfo.developer ?? "unknown").trim().toLowerCase().replace(/\s+/g, "-")
-}
-
-function getRawHandle(modelInfo: ModelInfo, namespace: string) {
-  const idParts = modelInfo.id.split("/")
-  const idHandle = idParts[idParts.length - 1]?.trim()
-  if (idHandle) {
-    return idHandle
-  }
-
-  const strippedName = stripNamespace(modelInfo.name, namespace)
-  return strippedName || modelInfo.name.trim()
-}
-
-function normalizeHandle(rawHandle: string) {
-  return rawHandle
-    .trim()
-    .toLowerCase()
-    .replace(/[_\s/]+/g, "-")
-    .replace(/(\d)-(?=\d(?:-|$))/g, "$1.")
-    .replace(/-+/g, "-")
-    .replace(/^-|-$/g, "")
-}
-
-function formatVersionDate(dateToken: string) {
-  if (!/^(?:19|20)\d{6}$/.test(dateToken)) {
-    return dateToken
-  }
-
-  return `${dateToken.slice(0, 4)}-${dateToken.slice(4, 6)}-${dateToken.slice(6, 8)}`
-}
-
-function titleCaseToken(token: string) {
-  if (!token) {
-    return token
-  }
-
-  const knownCase = TOKEN_CASE_MAP[token]
-  if (knownCase) {
-    return knownCase
-  }
-
-  if (/^\d+(\.\d+)?$/.test(token)) {
-    return token
-  }
-
-  if (/^\d+(\.\d+)?[bkmt]$/i.test(token)) {
-    return `${token.slice(0, -1)}${token.slice(-1).toUpperCase()}`
-  }
-
-  if (/^v\d/i.test(token)) {
-    return `v${token.slice(1)}`
-  }
-
-  return token.charAt(0).toUpperCase() + token.slice(1)
-}
-
-function humanizeHandle(handle: string) {
-  return handle
-    .split("-")
-    .filter(Boolean)
-    .map((token) => {
-      if (/^(?:19|20)\d{6}$/.test(token)) {
-        return formatVersionDate(token)
-      }
-
-      return titleCaseToken(token)
-    })
-    .join(" ")
-}
-
-function splitVersionParts(normalizedHandle: string) {
-  const match = normalizedHandle.match(/^(.*?)-((?:19|20)\d{6})(?:-(.+))?$/)
-
-  if (!match) {
-    return {
-      familySlug: normalizedHandle,
-      variantKey: "base",
-      variantLabel: "Current",
-      versionDate: undefined,
-      versionQualifier: undefined,
-    }
-  }
-
-  const [, baseSlug, dateToken, qualifier] = match
-  const qualifierLabel = qualifier ? humanizeHandle(qualifier) : null
-  const formattedDate = formatVersionDate(dateToken)
-
-  return {
-    familySlug: baseSlug,
-    variantKey: qualifier ? `${dateToken}-${qualifier}` : dateToken,
-    variantLabel: qualifierLabel ? `${formattedDate} · ${qualifierLabel}` : formattedDate,
-    versionDate: formattedDate,
-    versionQualifier: qualifierLabel ?? undefined,
-  }
-}
-
-export function getCanonicalModelIdentity(modelInfo: ModelInfo): ParsedModelIdentity {
-  const namespace = getNamespace(modelInfo)
-  const rawHandle = getRawHandle(modelInfo, namespace)
-  const normalizedHandle = normalizeHandle(rawHandle)
-  const versionParts = splitVersionParts(normalizedHandle)
-  const familyName = humanizeHandle(versionParts.familySlug)
-  const familyId = `${namespace}/${versionParts.familySlug}`
-  const variantDisplayName =
-    versionParts.variantKey === "base"
-      ? familyName
-      : `${familyName} (${versionParts.variantLabel})`
-
-  return {
-    namespace,
-    rawHandle,
-    normalizedHandle,
-    familySlug: versionParts.familySlug,
-    familyId,
-    familyName,
-    variantKey: versionParts.variantKey,
-    variantLabel: versionParts.variantLabel,
-    variantDisplayName,
-    versionDate: versionParts.versionDate,
-    versionQualifier: versionParts.versionQualifier,
-  }
-}
-
-export function getModelFamilyRouteId(model: ModelInfo | string) {
-  const familyId =
-    typeof model === "string" ? model.trim() : getCanonicalModelIdentity(model).familyId
-
-  // Use RFC 3986 percent-encoding (matches `models_view.route_id` /
-  // `model_route_id` produced by Stage J). The legacy `/` → `__`
-  // transform was a holdover from when model ids were embedded into
-  // HF-dataset paths and `/` was disallowed; the v2 route lookup
-  // matches `route_id` directly, so a `__`-formatted slug returns 404.
-  return encodeURIComponent(familyId)
-}

lib/model-url-redirects-build.ts

@@ -0,0 +1,120 @@
+// Redirect map builder (pure, testable).
+//
+// The producer's warehouse makes the GROUP the addressable model (its
+// `model_id` / `model_route_id`). Every variant, dated snapshot, or
+// older-cased spelling the producer folds into a group lives only in that group
+// row's `raw_model_ids` list and has no page of its own. This builds the
+// old-route -> canonical-group-route 301 map straight from `models_view`, so a
+// link to any folded spelling lands on the real group page. Every redirect
+// target is an addressable `model_route_id` by construction.
+//
+// Kept separate from the generated data file (`lib/model-url-redirects.ts`) so
+// the generator script AND unit tests can exercise the build logic without
+// importing the (large) generated map.
+//
+// Route id form: RFC 3986 percent-encoded whole id (`/` -> `%2F`), matching the
+// producer's `route_id` column and middleware's `encodeURIComponent` lookup key.
+
+export interface ModelsViewRow {
+  /** canonical encoded route for this group — the redirect TARGET. */
+  model_route_id: string
+  model_id?: string | null
+  route_id?: string | null
+  model_key?: string | null
+  model_group_id?: string | null
+  /** raw spellings folded into this group (HF-cased, may contain `/`). */
+  raw_model_ids?: string[] | null
+}
+
+/** Encode a plain model id into the producer `route_id` form. */
+export function routeId(id: string): string {
+  return encodeURIComponent(id.trim())
+}
+
+export interface BuildResult {
+  /** old encoded route id -> canonical (group) encoded route id. */
+  redirects: Map<string, string>
+  /**
+   * Raw spellings that fan out to MORE THAN ONE group (would be ambiguous to
+   * redirect). Excluded from the map and surfaced for review rather than
+   * picking one arbitrarily. Expected to be empty for a well-formed warehouse.
+   */
+  ambiguous: Map<string, Set<string>>
+}
+
+/**
+ * Build the redirect map from models_view rows. Pure and deterministic.
+ *
+ * A folded spelling is redirected to its owning group's `model_route_id` unless
+ * (a) it already equals that route, or (b) it is itself an addressable page
+ * (some row's own id) — we never hijack a working URL. A spelling that maps to
+ * multiple distinct groups is recorded as ambiguous and excluded.
+ */
+export function buildRedirectsFromModelsView(rows: ModelsViewRow[]): BuildResult {
+  // Pass 1: every route form that resolves to a real page (mirrors the
+  // getModelSummaryById lookup columns), so we never redirect a live URL.
+  const addressable = new Set<string>()
+  for (const r of rows) {
+    if (r.model_route_id) addressable.add(r.model_route_id)
+    if (r.route_id) addressable.add(r.route_id)
+    if (r.model_id) addressable.add(routeId(r.model_id))
+    if (r.model_group_id) addressable.add(routeId(r.model_group_id))
+    if (r.model_key) addressable.add(routeId(r.model_key))
+  }
+
+  // Pass 2: map each folded raw spelling to its owning group route.
+  const candidates = new Map<string, Set<string>>()
+  for (const r of rows) {
+    const target = r.model_route_id
+    if (!target) continue
+    for (const raw of r.raw_model_ids ?? []) {
+      if (!raw) continue
+      const old = routeId(raw)
+      if (old === target) continue // already the canonical route
+      if (addressable.has(old)) continue // has its own page; don't hijack it
+      const set = candidates.get(old) ?? new Set<string>()
+      set.add(target)
+      candidates.set(old, set)
+    }
+  }
+
+  const redirects = new Map<string, string>()
+  const ambiguous = new Map<string, Set<string>>()
+  for (const [old, targets] of candidates) {
+    if (targets.size === 1) redirects.set(old, [...targets][0])
+    else ambiguous.set(old, targets)
+  }
+  return { redirects, ambiguous }
+}
+
+/** Serialise a redirect map to the generated TS module source. */
+export function serializeRedirectModule(
+  redirects: Map<string, string>,
+  meta: { source: string },
+): string {
+  const sorted = [...redirects.entries()].sort(([a], [b]) => (a < b ? -1 : a > b ? 1 : 0))
+  const entries = sorted.map(([o, n]) => `  [${JSON.stringify(o)}, ${JSON.stringify(n)}],`).join("\n")
+  return `// GENERATED by scripts/generate-model-redirects.ts — do not edit by hand.
+//
+// Derived from the warehouse models_view (${meta.source}). Each entry maps a
+// folded model spelling (a variant / dated snapshot / older-cased id the
+// producer collapses into a group) to that group's canonical route, so a link
+// to any folded spelling 301s to the real group page. Every target is an
+// addressable model_route_id by construction.
+//
+// Entries: ${redirects.size}.
+//
+// Old encoded route id -> canonical encoded route id (RFC 3986, \`/\` -> %2F).
+// middleware.ts looks up the incoming route here and issues a 301 (query params
+// preserved).
+
+/** old encoded route id -> canonical (group) encoded route id. */
+export const MODEL_URL_REDIRECTS = new Map<string, string>([
+${entries}
+])
+
+export function resolveModelRedirect(routeId: string): string | undefined {
+  return MODEL_URL_REDIRECTS.get(routeId)
+}
+`
+}

lib/model-url-redirects.ts

@@ -0,0 +1,1128 @@
+// GENERATED by scripts/generate-model-redirects.ts — do not edit by hand.
+//
+// Derived from the warehouse models_view (https://huggingface.co/datasets/evaleval/card_backend/resolve/main/warehouse/2026-06-05T02-06-42Z). Each entry maps a
+// folded model spelling (a variant / dated snapshot / older-cased id the
+// producer collapses into a group) to that group's canonical route, so a link
+// to any folded spelling 301s to the real group page. Every target is an
+// addressable model_route_id by construction.
+//
+// Entries: 1107.
+//
+// Old encoded route id -> canonical encoded route id (RFC 3986, `/` -> %2F).
+// middleware.ts looks up the incoming route here and issues a 301 (query params
+// preserved).
+
+/** old encoded route id -> canonical (group) encoded route id. */
+export const MODEL_URL_REDIRECTS = new Map<string, string>([
+  ["01-ai%2FYi-1.5-34B-Chat", "01-ai%2FYi-1.5-34B"],
+  ["01-ai%2FYi-1.5-6B-Chat", "01-ai%2FYi-1.5-6B"],
+  ["01-ai%2FYi-1.5-9B-Chat", "01-ai%2FYi-1.5-9B"],
+  ["01-ai%2FYi-34B-Chat", "01-ai%2FYi-34B"],
+  ["01-ai%2FYi-6B-Chat", "01-ai%2FYi-6B"],
+  ["01-ai%2FYi-Coder-9B-Chat", "01-ai%2Fyi-coder-9b"],
+  ["01-ai%2Fyi-1.5-34b-chat", "01-ai%2FYi-1.5-34B"],
+  ["01-ai%2Fyi-1.5-6b-chat", "01-ai%2FYi-1.5-6B"],
+  ["01-ai%2Fyi-1.5-9b-chat", "01-ai%2FYi-1.5-9B"],
+  ["01-ai%2Fyi-34b", "01-ai%2FYi-34B"],
+  ["01-ai%2Fyi-34b-chat", "01-ai%2FYi-34B"],
+  ["01-ai%2Fyi-6b", "01-ai%2FYi-6B"],
+  ["01-ai%2Fyi-6b-chat", "01-ai%2FYi-6B"],
+  ["1TuanPham%2FT-VisStar-v0.1", "1TuanPham%2FT-VisStar-7B-v0.1"],
+  ["Ahjeong%2FMMPO_Gemma_7b", "Ahjeong%2FMMPO_Gemma_7b_gamma1.1_epoch3"],
+  ["Alepach%2FnotHumpback-M1", "Alepach%2FnotHumpback-M1-3b"],
+  ["Anthropic%2Fclaude-3-5-sonnet-20240620", "anthropic%2Fclaude-3.5-sonnet"],
+  ["Anthropic%2Fclaude-3-haiku-20240307", "anthropic%2Fclaude-3-haiku"],
+  ["Anthropic%2Fclaude-3-opus-20240229", "anthropic%2Fclaude-3-opus"],
+  ["Anthropic%2Fclaude-3-sonnet-20240229", "anthropic%2Fclaude-3-sonnet"],
+  ["Anthropic-LM-v4-s3-52B", "anthropic-lm-v4-s3-52b"],
+  ["ArliAI%2FArliAI-RPMax-12B-v1.1", "ArliAI%2FMistral-Nemo-12B-ArliAI-RPMax-v1.1"],
+  ["AtlaAI%2FSelene-1", "AtlaAI%2FSelene-1-Llama-3.3-70B"],
+  ["BAAI%2FInfinity-Instruct-7M-0729-Llama3_1-8B", "BAAI%2FInfinity-Instruct-7M-Gen-Llama3_1-8B"],
+  ["BAAI%2FInfinity-Instruct-7M-0729-mistral-7B", "BAAI%2FInfinity-Instruct-7M-Gen-mistral-7B"],
+  ["BEE-spoke-data%2FtFINE-900m-e16-d32-flan-infinity-instruct-7m-T2T_en-1024", "BEE-spoke-data%2FtFINE-900m-e16-d32-instruct"],
+  ["CYFRAGOVPL%2FLlama-PLLuM-8B-base", "CYFRAGOVPL%2FLlama-PLLuM-8B-base-2412"],
+  ["CYFRAGOVPL%2FLlama-PLLuM-8B-chat", "CYFRAGOVPL%2FLlama-PLLuM-8B-chat-2412"],
+  ["CYFRAGOVPL%2FPLLuM-12B-base", "CYFRAGOVPL%2FPLLuM-12B-base-2412"],
+  ["CYFRAGOVPL%2FPLLuM-12B-chat", "CYFRAGOVPL%2FPLLuM-12B-chat-2412"],
+  ["CYFRAGOVPL%2FPLLuM-12B-nc-base", "CYFRAGOVPL%2FPLLuM-12B-nc-base-2412"],
+  ["CYFRAGOVPL%2FPLLuM-12B-nc-chat", "CYFRAGOVPL%2FPLLuM-12B-nc-chat-2412"],
+  ["Cohere%20March%202024", "cohere-march-2024"],
+  ["Cohere%20May%202024", "cohere-may-2024"],
+  ["CohereForAI%2Faya-23-35B", "CohereLabs%2Faya-23-35B"],
+  ["CohereForAI%2Faya-23-8B", "CohereLabs%2Faya-23-8B"],
+  ["CohereForAI%2Faya-expanse-32b", "CohereLabs%2Faya-expanse-32b"],
+  ["CohereForAI%2Faya-expanse-8b", "CohereLabs%2Faya-expanse-8b"],
+  ["CohereForAI%2Fc4ai-command-r-plus", "CohereLabs%2Fc4ai-command-r-plus"],
+  ["CohereForAI%2Fc4ai-command-r-plus-08-2024", "CohereLabs%2Fc4ai-command-r-plus-08-2024"],
+  ["CohereForAI%2Fc4ai-command-r-v01", "CohereLabs%2Fc4ai-command-r-v01"],
+  ["CohereForAI%2Fc4ai-command-r7b-12-2024", "CohereLabs%2Fc4ai-command-r7b-12-2024"],
+  ["CohereLabs%2Fcommand-a-reasoning-08-2025", "cohere%2Fcommand-a-reasoning"],
+  ["DeepAutoAI%2FExplore_Llama-3.1-8B-Inst", "DeepAuto-AI%2FExplore_Llama-3.1-8B-Inst"],
+  ["DeepAutoAI%2FExplore_Llama-3.2-1B-Inst", "DeepAuto-AI%2FExplore_Llama-3.2-1B-Inst"],
+  ["DeepAutoAI%2FExplore_Llama-3.2-1B-Inst_v0", "DeepAuto-AI%2FExplore_Llama-3.2-1B-Inst_v0"],
+  ["DeepAutoAI%2FExplore_Llama-3.2-1B-Inst_v1", "DeepAuto-AI%2FExplore_Llama-3.2-1B-Inst_v1"],
+  ["DeepAutoAI%2FExplore_Llama-3.2-1B-Inst_v1.1", "DeepAuto-AI%2FExplore_Llama-3.2-1B-Inst_v1.1"],
+  ["DeepAutoAI%2Fcausal_gpt2", "DeepAuto-AI%2Fcausal_gpt2"],
+  ["DeepAutoAI%2Fd2nwg_Llama-3.1-8B-Instruct-v0.0", "DeepAuto-AI%2Fd2nwg_Llama-3.1-8B-Instruct-v0.0"],
+  ["DeepAutoAI%2Fd2nwg_causal_gpt2", "DeepAuto-AI%2Fd2nwg_causal_gpt2"],
+  ["DeepAutoAI%2Fd2nwg_causal_gpt2_v1", "DeepAuto-AI%2Fd2nwg_causal_gpt2_v1"],
+  ["DeepAutoAI%2Fldm_soup_Llama-3.1-8B-Inst", "DeepAuto-AI%2Fldm_soup_Llama-3.1-8B-Inst"],
+  ["DeepAutoAI%2Fldm_soup_Llama-3.1-8B-Instruct-v0.0", "DeepAuto-AI%2Fldm_soup_Llama-3.1-8B-Instruct-v0.0"],
+  ["DeepAutoAI%2Fldm_soup_Llama-3.1-8B-Instruct-v0.1", "DeepAuto-AI%2Fldm_soup_Llama-3.1-8B-Instruct-v0.1"],
+  ["DeepMount00%2FLexora-Lite-3B", "DeepMount00%2FLexora-Lite-3B_v2"],
+  ["DeepMount00%2FLlama-3.1-8b-Ita", "DeepMount00%2FLlama-3.1-8b-ITA"],
+  ["DeepSeek-Coder-V2-Lite-Instruct", "deepseek-coder-v2-lite-instruct"],
+  ["DeepSeek-V2-Lite-Chat", "deepseek-ai%2FDeepSeek-V2-Lite"],
+  ["Dracarys2-72B-Instruct", "dracarys2-72b-instruct"],
+  ["HuggingFaceTB%2FSmolLM-1.7B-Instruct", "HuggingFaceTB%2FSmolLM-1.7B"],
+  ["HuggingFaceTB%2FSmolLM-135M-Instruct", "HuggingFaceTB%2FSmolLM-135M"],
+  ["HuggingFaceTB%2FSmolLM-360M-Instruct", "HuggingFaceTB%2FSmolLM-360M"],
+  ["HuggingFaceTB%2FSmolLM2-1.7B-Instruct", "HuggingFaceTB%2FSmolLM2-1.7B"],
+  ["HuggingFaceTB%2FSmolLM2-135M-Instruct", "HuggingFaceTB%2FSmolLM2-135M"],
+  ["HuggingFaceTB%2FSmolLM2-360M-Instruct", "HuggingFaceTB%2FSmolLM2-360M"],
+  ["Krystalan%2FDRT-o1-14B", "Krystalan%2FDRT-14B"],
+  ["Krystalan%2FDRT-o1-7B", "Krystalan%2FDRT-7B"],
+  ["LEESM%2Fllama-2-7b-hf-lora-oki100p", "leesm%2Fllama-2-7b-hf-lora-oki100p"],
+  ["LEESM%2Fllama-2-7b-hf-lora-oki10p", "leesm%2Fllama-2-7b-hf-lora-oki10p"],
+  ["LEESM%2Fllama-3-8b-bnb-4b-kowiki231101", "leesm%2Fllama-3-8b-bnb-4b-kowiki231101"],
+  ["LEESM%2Fllama-3-Korean-Bllossom-8B-trexlab-oki10p", "leesm%2Fllama-3-Korean-Bllossom-8B-trexlab-oki10p"],
+  ["Llama-2-7b-chat-hf", "meta-llama%2FLlama-2-7b-chat-hf"],
+  ["Llama-3.1-Nemotron-70B-Instruct-HF", "nvidia%2FLlama-3.1-Nemotron-70B-Instruct-HF"],
+  ["Meta-Llama-3-70B-Instruct", "meta-llama%2Fllama-3"],
+  ["Meta-Llama-3-8B-Instruct", "meta-llama%2Fllama-3"],
+  ["Mistral-7B-Instruct-v0.2", "mistralai%2FMistral-7B-Instruct-v0.2"],
+  ["Mistral-7B-Instruct-v0.3", "mistralai%2Fmistral-7b"],
+  ["Mixtral-8x22B-Instruct-v0.1", "mistralai%2FMixtral-8x22B-Instruct-v0.1"],
+  ["Mixtral-8x7B-Instruct-v0.1", "mistralai%2FMixtral-8x7B-Instruct-v0.1"],
+  ["OpenAssistant%2Foasst-rm-2-pythia-6.9b-epoch-1", "OpenAssistant%2Foasst-rm-pythia"],
+  ["OpenAssistant%2Foasst-rm-2.1-pythia-1.4b-epoch-2.5", "OpenAssistant%2Foasst-rm-pythia"],
+  ["OpenHermes-2.5-Mistral-7B", "teknium%2FOpenHermes-2.5-Mistral-7B"],
+  ["Phi-3-medium-128k-instruct", "microsoft%2Fphi-3-medium-128k"],
+  ["Phi-3-medium-4k-instruct", "microsoft%2Fphi-3-medium-4k"],
+  ["Phi-3-mini-128k-instruct", "microsoft%2Fphi-3-mini-128k"],
+  ["Phi-3-mini-4k-instruct", "microsoft%2Fphi-3-mini-4k"],
+  ["Phi-3-small-128k-instruct", "microsoft%2Fphi-3-small-128k"],
+  ["Phi-3-small-8k-instruct", "microsoft%2Fphi-3-small-8k"],
+  ["Phi-3.5-MoE-instruct", "microsoft%2Fphi-3.5-moe"],
+  ["Phi-3.5-mini-instruct", "microsoft%2Fphi-3.5-mini"],
+  ["Quazim0t0%2FODB-14b-sce", "Quazim0t0%2FODB-14B-sce"],
+  ["Qwen%2FQwen1.5-110B-Chat", "Qwen%2FQwen1.5-110B"],
+  ["Qwen%2FQwen1.5-14B-Chat", "Qwen%2FQwen1.5-14B"],
+  ["Qwen%2FQwen1.5-32B-Chat", "Qwen%2FQwen1.5-32B"],
+  ["Qwen%2FQwen1.5-72B-Chat", "Qwen%2FQwen1.5-72B"],
+  ["Qwen%2FQwen1.5-7B-Chat", "Qwen%2FQwen1.5-7B"],
+  ["Qwen%2FQwen2-1.5B-Instruct", "Qwen%2FQwen2-1.5B"],
+  ["Qwen%2FQwen2-72B-Instruct", "Qwen%2FQwen2-72B"],
+  ["Qwen%2FQwen2-7B-Instruct", "Qwen%2FQwen2-7B"],
+  ["Qwen%2FQwen2-VL-2B-Instruct", "qwen%2Fqwen2-vl-2b"],
+  ["Qwen%2FQwen2-VL-72B-Instruct", "Qwen%2FQwen2-VL-72B"],
+  ["Qwen%2FQwen2-VL-7B-Instruct", "qwen%2Fqwen2-vl-7b"],
+  ["Qwen%2FQwen2.5-0.5B-Instruct", "Qwen%2FQwen2.5-0.5B"],
+  ["Qwen%2FQwen2.5-1.5B-Instruct", "Qwen%2FQwen2.5-1.5B"],
+  ["Qwen%2FQwen2.5-14B-Instruct", "Qwen%2FQwen2.5-14B"],
+  ["Qwen%2FQwen2.5-32B-Instruct", "Qwen%2FQwen2.5-32B"],
+  ["Qwen%2FQwen2.5-3B-Instruct", "Qwen%2FQwen2.5-3B"],
+  ["Qwen%2FQwen2.5-72B-Instruct", "Qwen%2FQwen2.5-72B"],
+  ["Qwen%2FQwen2.5-72B-Instruct-Turbo", "Qwen%2FQwen2.5-72B"],
+  ["Qwen%2FQwen2.5-7B-Instruct", "Qwen%2FQwen2.5-7B"],
+  ["Qwen%2FQwen2.5-7B-Instruct-Turbo", "Qwen%2FQwen2.5-7B"],
+  ["Qwen%2FQwen2.5-Coder-14B-Instruct", "Qwen%2FQwen2.5-Coder-14B"],
+  ["Qwen%2FQwen2.5-Coder-32B-Instruct", "Qwen%2FQwen2.5-Coder-32B"],
+  ["Qwen%2FQwen2.5-Coder-7B-Instruct", "Qwen%2FQwen2.5-Coder-7B"],
+  ["Qwen%2FQwen2.5-Math-72B-Instruct", "qwen%2Fqwen2.5-math-72b"],
+  ["Qwen%2FQwen2.5-Math-7B-Instruct", "Qwen%2FQwen2.5-Math-7B"],
+  ["Qwen%2FQwen2.5-VL-72B-Instruct", "alibaba%2Fqwen2.5-vl-72b"],
+  ["Qwen%2FQwen3-30B-A3B-Thinking-2507", "Qwen%2FQwen3-30B-A3B"],
+  ["Qwen%2FQwen3-Next-80B-A3B-Thinking", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["Qwen%2FQwen3-VL-235B-A22B-Thinking-FP8", "alibaba%2Fqwen3-vl-235b-a22b"],
+  ["Qwen%2FQwen3.5-122B-A10B", "alibaba%2Fqwen3-5-122b"],
+  ["Qwen%2FQwen3.5-35B-A3B", "qwen%2Fqwen3.5-35b"],
+  ["Qwen%2FQwen3.5-397B-A17B", "alibaba%2Fqwen3-5-397b"],
+  ["Qwen%2FQwen3.6-35B-A3B", "alibaba%2Fqwen3-6-35b"],
+  ["Qwen1.5-0.5B-Chat", "Qwen%2FQwen1.5-0.5B-Chat"],
+  ["Qwen1.5-1.8B-Chat", "Qwen%2FQwen1.5-1.8B-Chat"],
+  ["Qwen1.5-110B-Chat", "Qwen%2FQwen1.5-110B"],
+  ["Qwen1.5-4B-Chat", "Qwen%2FQwen1.5-4B-Chat"],
+  ["Qwen1.5-72B-Chat", "Qwen%2FQwen1.5-72B"],
+  ["Qwen1.5-7B-Chat", "Qwen%2FQwen1.5-7B"],
+  ["Qwen2-0.5B-Instruct", "Qwen%2FQwen2-0.5B-Instruct"],
+  ["Qwen2-1.5B-Instruct", "Qwen%2FQwen2-1.5B"],
+  ["Qwen2-72B-Instruct", "Qwen%2FQwen2-72B"],
+  ["Qwen2-7B-Instruct", "Qwen%2FQwen2-7B"],
+  ["Ray2333%2FGRM-llama3.2-3B-rewardmodel-ft", "Ray2333%2FGRM-Llama3.2-3B-rewardmodel-ft"],
+  ["Reflection-Llama-3.1-70B", "mattshumer%2FReflection-Llama-3.1-70B"],
+  ["Sakalti%2FMagro-7B-v1.1", "Sakalti%2FMagro-7b-v1.1"],
+  ["Sakalti%2FOxyge1-33B", "Sakalti%2Foxyge1-33B"],
+  ["Sakalti%2Flight-3b-beta", "Sakalti%2Flight-3B-beta"],
+  ["Sakalti%2Fllama-3-yanyuedao-8b-instruct", "Sakalti%2Fllama-3-yanyuedao-8b-Instruct"],
+  ["SkyOrbis%2FSKY-Ko-Qwen2.5-3B-Instruct", "SkyOrbis%2FSKY-Ko-Qwen2.5-3B-Instruct-SFT"],
+  ["Skywork%2FSkywork-Reward-V2-Qwen3-0.6B", "Skywork%2Fskywork-reward-v2-qwen3"],
+  ["Skywork%2FSkywork-Reward-V2-Qwen3-1.7B", "Skywork%2Fskywork-reward-v2-qwen3"],
+  ["Skywork%2FSkywork-Reward-V2-Qwen3-4B", "Skywork%2Fskywork-reward-v2-qwen3"],
+  ["Skywork%2FSkywork-Reward-V2-Qwen3-8B", "Skywork%2Fskywork-reward-v2-qwen3"],
+  ["Smaug-Qwen2-72B-Instruct", "abacusai%2FSmaug-Qwen2-72B-Instruct"],
+  ["Solshine%2FBrimful-merged-replete", "Solshine%2FBrimful-12B-merged-replete"],
+  ["Starling-LM-7B-beta", "starling-lm-7b-beta"],
+  ["SultanR%2FSmolTulu-1.7b-it-v0", "SultanR%2FSmolTulu-1.7b-Instruct"],
+  ["THUDM%2Fcogvlm2-llama3-chat-19B", "zai-org%2Fcogvlm2-llama3-chat-19B"],
+  ["THUDM%2Fglm-4-9b", "zai-org%2Fglm-4-9b"],
+  ["THUDM%2Fglm-4-9b-chat", "zai-org%2Fglm-4-9b"],
+  ["THUDM%2Fglm-4-9b-chat-1m", "zai-org%2Fglm-4-9b-chat-1m"],
+  ["THUDM%2Fglm-4-9b-chat-1m-hf", "zai-org%2Fglm-4-9b-chat-1m-hf"],
+  ["THUDM%2Fglm-4-9b-chat-hf", "zai-org%2Fglm-4-9b-chat-hf"],
+  ["XiaomiMiMo%2FMiMo-V2-Flash", "xiaomi%2Fmimo-v2"],
+  ["YOYO-AI%2FQwen2.5-14B-1M-YOYO-V3", "YOYO-AI%2FQwen2.5-14B-YOYO-V3"],
+  ["Yi-6B-Chat", "01-ai%2FYi-6B"],
+  ["acm_rewrite_qwen2-72B-Chat", "acm-rewrite-qwen2-72b-chat"],
+  ["ai2%2Folmo-3-1-32b-instruct", "allenai%2FOlmo-3.1-32B-Instruct"],
+  ["ai2%2Folmo-3-1-32b-think", "allenai%2FOlmo-3.1-32B-Think"],
+  ["ai21%2Fjamba-1.5-large", "ai21-labs%2Fai21-jamba-1-5-large"],
+  ["ai21%2Fjamba-1.5-mini", "ai21-labs%2Fai21-jamba-1-5-mini"],
+  ["ai21%2Fjamba-instruct", "ai21%2Fjamba"],
+  ["alibaba%2FQwen2.5-72B-Instruct", "Qwen%2FQwen2.5-72B"],
+  ["alibaba%2FQwen2.5-Coder-32B-Instruct", "Qwen%2FQwen2.5-Coder-32B"],
+  ["alibaba%2FQwen3-Coder-30B-A3B-Instruct", "alibaba%2Fqwen3-coder-30b-a3b"],
+  ["alibaba%2Fqwen-1.5-0.5b-chat", "Qwen%2FQwen1.5-0.5B-Chat"],
+  ["alibaba%2Fqwen-1.5-1.8b-chat", "Qwen%2FQwen1.5-1.8B-Chat"],
+  ["alibaba%2Fqwen-1.5-14b-chat", "Qwen%2FQwen1.5-14B"],
+  ["alibaba%2Fqwen-1.5-32b-chat", "Qwen%2FQwen1.5-32B"],
+  ["alibaba%2Fqwen-1.5-4b-chat", "Qwen%2FQwen1.5-4B-Chat"],
+  ["alibaba%2Fqwen-1.5-7b-chat", "Qwen%2FQwen1.5-7B"],
+  ["alibaba%2Fqwen-2-1.5b-instruct", "Qwen%2FQwen2-1.5B"],
+  ["alibaba%2Fqwen-2-72b-instruct", "Qwen%2FQwen2-72B"],
+  ["alibaba%2Fqwen-2-7b-instruct", "Qwen%2FQwen2-7B"],
+  ["alibaba%2Fqwen-2.5-0.5b-instruct", "Qwen%2FQwen2.5-0.5B"],
+  ["alibaba%2Fqwen-2.5-1.5b-instruct", "Qwen%2FQwen2.5-1.5B"],
+  ["alibaba%2Fqwen-2.5-14b-instruct", "Qwen%2FQwen2.5-14B"],
+  ["alibaba%2Fqwen-2.5-32b-instruct", "Qwen%2FQwen2.5-32B"],
+  ["alibaba%2Fqwen-2.5-3b-instruct", "Qwen%2FQwen2.5-3B"],
+  ["alibaba%2Fqwen-2.5-72b-instruct", "Qwen%2FQwen2.5-72B"],
+  ["alibaba%2Fqwen-2.5-7b-instruct", "Qwen%2FQwen2.5-7B"],
+  ["alibaba%2Fqwen1.5-110b", "Qwen%2FQwen1.5-110B"],
+  ["alibaba%2Fqwen1.5-110b-chat", "Qwen%2FQwen1.5-110B"],
+  ["alibaba%2Fqwen1.5-14b-chat", "Qwen%2FQwen1.5-14B"],
+  ["alibaba%2Fqwen1.5-72b-chat", "Qwen%2FQwen1.5-72B"],
+  ["alibaba%2Fqwen1.5-7b-chat", "Qwen%2FQwen1.5-7B"],
+  ["alibaba%2Fqwen2-0.5b", "Qwen%2FQwen2-0.5B"],
+  ["alibaba%2Fqwen2-0.5b-instruct", "Qwen%2FQwen2-0.5B-Instruct"],
+  ["alibaba%2Fqwen2-1.5b", "Qwen%2FQwen2-1.5B"],
+  ["alibaba%2Fqwen2-1.5b-instruct", "Qwen%2FQwen2-1.5B"],
+  ["alibaba%2Fqwen2-5-72b-instruct", "Qwen%2FQwen2.5-72B"],
+  ["alibaba%2Fqwen2-5-coder-32b-instruct", "Qwen%2FQwen2.5-Coder-32B"],
+  ["alibaba%2Fqwen2-5-coder-7b-instruct", "Qwen%2FQwen2.5-Coder-7B"],
+  ["alibaba%2Fqwen2-72b-instruct", "Qwen%2FQwen2-72B"],
+  ["alibaba%2Fqwen2-7b", "Qwen%2FQwen2-7B"],
+  ["alibaba%2Fqwen2-7b-instruct", "Qwen%2FQwen2-7B"],
+  ["alibaba%2Fqwen2.5-0.5b", "Qwen%2FQwen2.5-0.5B"],
+  ["alibaba%2Fqwen2.5-1.5b", "Qwen%2FQwen2.5-1.5B"],
+  ["alibaba%2Fqwen2.5-14b", "Qwen%2FQwen2.5-14B"],
+  ["alibaba%2Fqwen2.5-32b", "Qwen%2FQwen2.5-32B"],
+  ["alibaba%2Fqwen2.5-32b-instruct", "Qwen%2FQwen2.5-32B"],
+  ["alibaba%2Fqwen2.5-3b", "Qwen%2FQwen2.5-3B"],
+  ["alibaba%2Fqwen2.5-72b", "Qwen%2FQwen2.5-72B"],
+  ["alibaba%2Fqwen2.5-7b", "Qwen%2FQwen2.5-7B"],
+  ["alibaba%2Fqwen3-0.6b-instruct", "qwen%2Fqwen3-0.6b"],
+  ["alibaba%2Fqwen3-0.6b-instruct-reasoning", "qwen%2Fqwen3-0.6b"],
+  ["alibaba%2Fqwen3-1.7b-instruct", "Qwen%2FQwen3-1.7B"],
+  ["alibaba%2Fqwen3-1.7b-instruct-reasoning", "Qwen%2FQwen3-1.7B"],
+  ["alibaba%2Fqwen3-14b-instruct", "Qwen%2FQwen3-14B"],
+  ["alibaba%2Fqwen3-14b-instruct-reasoning", "Qwen%2FQwen3-14B"],
+  ["alibaba%2Fqwen3-235b-a22b", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-235b-a22b-instruct", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-235b-a22b-instruct-2507", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-235b-a22b-instruct-2507-reasoning", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-235b-a22b-instruct-reasoning", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-235b-a22b-thinking-2507", "Qwen%2FQwen3-235B-A22B"],
+  ["alibaba%2Fqwen3-30b-a3b", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-2507", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-2507-reasoning", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-base", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-instruct", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-instruct-reasoning", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-thinking", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-30b-a3b-thinking-2507", "Qwen%2FQwen3-30B-A3B"],
+  ["alibaba%2Fqwen3-32b-instruct", "Qwen%2FQwen3-32B"],
+  ["alibaba%2Fqwen3-32b-instruct-reasoning", "Qwen%2FQwen3-32B"],
+  ["alibaba%2Fqwen3-4b-2507-instruct", "alibaba%2Fqwen3-4b"],
+  ["alibaba%2Fqwen3-4b-2507-instruct-reasoning", "alibaba%2Fqwen3-4b"],
+  ["alibaba%2Fqwen3-4b-instruct-reasoning", "alibaba%2Fqwen3-4b"],
+  ["alibaba%2Fqwen3-5-0-8b", "Qwen%2FQwen3.5-0.8B"],
+  ["alibaba%2Fqwen3-5-122b-a10b", "alibaba%2Fqwen3-5-122b"],
+  ["alibaba%2Fqwen3-5-27b", "Qwen%2FQwen3.5-27B"],
+  ["alibaba%2Fqwen3-5-2b", "Qwen%2FQwen3.5-2B"],
+  ["alibaba%2Fqwen3-5-35b-a3b", "qwen%2Fqwen3.5-35b"],
+  ["alibaba%2Fqwen3-5-397b-a17b", "alibaba%2Fqwen3-5-397b"],
+  ["alibaba%2Fqwen3-5-4b", "Qwen%2FQwen3.5-4B"],
+  ["alibaba%2Fqwen3-5-9b", "Qwen%2FQwen3.5-9B"],
+  ["alibaba%2Fqwen3-6-plus", "alibaba%2Fqwen3.6-plus"],
+  ["alibaba%2Fqwen3-8b-instruct", "Qwen%2FQwen3-8B"],
+  ["alibaba%2Fqwen3-8b-instruct-reasoning", "Qwen%2FQwen3-8B"],
+  ["alibaba%2Fqwen3-coder-30b-a3b-instruct", "alibaba%2Fqwen3-coder-30b-a3b"],
+  ["alibaba%2Fqwen3-coder-480b-a35b-instruct", "alibaba%2Fqwen3-coder-480b-a35b"],
+  ["alibaba%2Fqwen3-max-2026-01-23", "alibaba%2Fqwen3-max"],
+  ["alibaba%2Fqwen3-next-80b-a3b-instruct", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["alibaba%2Fqwen3-next-80b-a3b-reasoning", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["alibaba%2Fqwen3-next-80b-a3b-thinking", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["alibaba%2Fqwen3-vl-235b-a22b-instruct", "alibaba%2Fqwen3-vl-235b-a22b"],
+  ["alibaba%2Fqwen3-vl-235b-a22b-reasoning", "alibaba%2Fqwen3-vl-235b-a22b"],
+  ["alibaba%2Fqwen3-vl-30b-a3b-instruct", "alibaba%2Fqwen3-vl-30b-a3b"],
+  ["alibaba%2Fqwen3-vl-30b-a3b-reasoning", "alibaba%2Fqwen3-vl-30b-a3b"],
+  ["alibaba%2Fqwen3-vl-32b-instruct", "Qwen%2FQwen3-VL-32B-Thinking"],
+  ["alibaba%2Fqwen3-vl-32b-reasoning", "Qwen%2FQwen3-VL-32B-Thinking"],
+  ["alibaba%2Fqwen3-vl-4b-instruct", "alibaba%2Fqwen3-vl-4b"],
+  ["alibaba%2Fqwen3-vl-4b-reasoning", "alibaba%2Fqwen3-vl-4b"],
+  ["alibaba%2Fqwen3-vl-8b-instruct", "Qwen%2FQwen3-VL-8B-Thinking"],
+  ["alibaba%2Fqwen3-vl-8b-reasoning", "Qwen%2FQwen3-VL-8B-Thinking"],
+  ["alibaba%2Fqwen3.5-0.8b", "Qwen%2FQwen3.5-0.8B"],
+  ["alibaba%2Fqwen3.5-122b-a10b", "alibaba%2Fqwen3-5-122b"],
+  ["alibaba%2Fqwen3.5-27b", "Qwen%2FQwen3.5-27B"],
+  ["alibaba%2Fqwen3.5-2b", "Qwen%2FQwen3.5-2B"],
+  ["alibaba%2Fqwen3.5-35b-a3b", "qwen%2Fqwen3.5-35b"],
+  ["alibaba%2Fqwen3.5-397b-a17b", "alibaba%2Fqwen3-5-397b"],
+  ["alibaba%2Fqwen3.5-4b", "Qwen%2FQwen3.5-4B"],
+  ["alibaba%2Fqwen3.5-9b", "Qwen%2FQwen3.5-9B"],
+  ["alibaba%2Fqwen3.5-flash", "alibaba%2Fqwen3-5-flash"],
+  ["alibaba%2Fqwen3.5-plus-thinking", "alibaba%2Fqwen3.5-plus"],
+  ["alibaba%2Fqwq-32b", "Qwen%2FQwQ-32B"],
+  ["alibaba%2Fqwq-32b-preview", "Qwen%2FQwQ-32B-Preview"],
+  ["allenai%2FOLMo-1.7-7B-hf", "allenai%2FOLMo-7B-hf"],
+  ["allenai%2FOLMo-2-1124-7B-Instruct", "allenai%2FOLMo-2-1124-7B"],
+  ["allenai%2FOLMo-7B-Instruct", "allenai%2FOLMo-7B"],
+  ["allenai%2FOLMoE-1B-7B-0125-Instruct", "allenai%2Folmoe-1b-7b-0125"],
+  ["allenai%2Folmo-1.7-7b", "allenai%2FOLMo-7B"],
+  ["allenai%2Folmo-2-0325-32b-instruct", "allenai%2Folmo-2-0325-32b"],
+  ["allenai%2Folmo-2-1124-13b-instruct", "allenai%2FOLMo-2-1124-13B"],
+  ["allenai%2Folmo-2-1124-7b-instruct", "allenai%2FOLMo-2-1124-7B"],
+  ["allenai%2Folmo-3.1-32b-instruct", "allenai%2FOlmo-3.1-32B-Instruct"],
+  ["allenai%2Folmo-7b", "allenai%2FOLMo-7B"],
+  ["allenai%2Folmoe-1b-7b-0125-instruct", "allenai%2Folmoe-1b-7b-0125"],
+  ["allknowingroger%2FGemmaSlerp-9B", "allknowingroger%2FGemmaslerp-9B"],
+  ["allknowingroger%2FGemmaSlerp2-9B", "allknowingroger%2FGemmaslerp2-9B"],
+  ["allknowingroger%2FGemmaSlerp4-10B", "allknowingroger%2FGemmaslerp4-10B"],
+  ["allknowingroger%2FQwenSlerp4-14B", "allknowingroger%2FQwenslerp4-14B"],
+  ["amazon%2FMegaBeam-Mistral-7B-300k", "aws-prototyping%2FMegaBeam-Mistral-7B-300k"],
+  ["anthropic%2Fclaude-3-5-haiku-20241022", "anthropic%2Fclaude-3-5-haiku"],
+  ["anthropic%2Fclaude-3-5-sonnet-20240620", "anthropic%2Fclaude-3.5-sonnet"],
+  ["anthropic%2Fclaude-3-5-sonnet-20241022", "anthropic%2Fclaude-3.5-sonnet"],
+  ["anthropic%2Fclaude-3-7-sonnet-20250219", "anthropic%2FClaude-3.7-Sonnet"],
+  ["anthropic%2Fclaude-3-haiku-20240307", "anthropic%2Fclaude-3-haiku"],
+  ["anthropic%2Fclaude-3-opus-20240229", "anthropic%2Fclaude-3-opus"],
+  ["anthropic%2Fclaude-3-sonnet-20240229", "anthropic%2Fclaude-3-sonnet"],
+  ["anthropic%2Fclaude-3.7-sonnet", "anthropic%2FClaude-3.7-Sonnet"],
+  ["anthropic%2Fclaude-3.7-sonnet-thinking", "anthropic%2FClaude-3.7-Sonnet"],
+  ["anthropic%2Fclaude-35-sonnet", "anthropic%2Fclaude-3.5-sonnet"],
+  ["anthropic%2Fclaude-4-sonnet", "anthropic%2FClaude-4-Sonnet"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-fc", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-prompt", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-thinking", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-thinking-16k", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-thinking-1k", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-thinking-32k", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-haiku-4-5-20251001-thinking-8k", "anthropic%2Fclaude-haiku-4.5"],
+  ["anthropic%2Fclaude-opus-4-1", "anthropic%2Fclaude-opus-4.1"],
+  ["anthropic%2Fclaude-opus-4-1-20250805", "anthropic%2Fclaude-opus-4.1"],
+  ["anthropic%2Fclaude-opus-4-20250514", "anthropic%2Fclaude-opus-4"],
+  ["anthropic%2Fclaude-opus-4-20250514-thinking-10k", "anthropic%2Fclaude-opus-4-20250514-thinking"],
+  ["anthropic%2Fclaude-opus-4-20250514-thinking-16k", "anthropic%2Fclaude-opus-4-20250514-thinking"],
+  ["anthropic%2Fclaude-opus-4-20250514-thinking-1k", "anthropic%2Fclaude-opus-4-20250514-thinking"],
+  ["anthropic%2Fclaude-opus-4-20250514-thinking-8k", "anthropic%2Fclaude-opus-4-20250514-thinking"],
+  ["anthropic%2Fclaude-opus-4-5", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-5-20251101", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-fc", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-prompt", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-thinking-16k", "anthropic%2Fclaude-opus-4-5-20251101-thinking"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-thinking-32k", "anthropic%2Fclaude-opus-4-5-20251101-thinking"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-thinking-64k", "anthropic%2Fclaude-opus-4-5-20251101-thinking"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-thinking-8k", "anthropic%2Fclaude-opus-4-5-20251101-thinking"],
+  ["anthropic%2Fclaude-opus-4-5-20251101-thinking-none", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-5-thinking", "anthropic%2Fclaude-opus-4.5"],
+  ["anthropic%2Fclaude-opus-4-6", "anthropic%2Fclaude-opus-4.6"],
+  ["anthropic%2Fclaude-opus-4-6-thinking", "anthropic%2Fclaude-opus-4.6"],
+  ["anthropic%2Fclaude-opus-4-6-thinking-120k-high", "anthropic%2Fclaude-opus-4.6"],
+  ["anthropic%2Fclaude-opus-4-6-thinking-120k-low", "anthropic%2Fclaude-opus-4.6"],
+  ["anthropic%2Fclaude-opus-4-6-thinking-120k-medium", "anthropic%2Fclaude-opus-4.6"],
+  ["anthropic%2Fclaude-opus-4-7", "anthropic%2Fclaude-opus-4.7"],
+  ["anthropic%2Fclaude-opus-4-thinking", "anthropic%2Fclaude-opus-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514-thinking", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514-thinking-10k", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514-thinking-16k-bedrock", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514-thinking-1k", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-20250514-thinking-8k-bedrock", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4-5", "anthropic%2Fclaude-sonnet-4.5"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929", "anthropic%2Fclaude-sonnet-4.5"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-fc", "anthropic%2Fclaude-sonnet-4.5"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-prompt", "anthropic%2Fclaude-sonnet-4.5"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-thinking-16k", "anthropic%2Fclaude-sonnet-4-5-20250929-thinking"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-thinking-1k", "anthropic%2Fclaude-sonnet-4-5-20250929-thinking"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-thinking-32k", "anthropic%2Fclaude-sonnet-4-5-20250929-thinking"],
+  ["anthropic%2Fclaude-sonnet-4-5-20250929-thinking-8k", "anthropic%2Fclaude-sonnet-4-5-20250929-thinking"],
+  ["anthropic%2Fclaude-sonnet-4-6", "anthropic%2Fclaude-sonnet-4.6"],
+  ["anthropic%2Fclaude-sonnet-4-6-high", "anthropic%2Fclaude-sonnet-4.6"],
+  ["anthropic%2Fclaude-sonnet-4-thinking", "anthropic%2Fclaude-sonnet-4"],
+  ["anthropic%2Fclaude-sonnet-4.6-high", "anthropic%2Fclaude-sonnet-4.6"],
+  ["arcee-ai%2Ftrinity-large-thinking", "arcee-ai%2FTrinity-Large-Thinking"],
+  ["avemio%2FGRAG-NEMO-12B-ORPO-HESSIAN-AI", "avemio%2FGerman-RAG-NEMO-12B-ORPO-HESSIAN-AI"],
+  ["bytedance%2Fseed-2.0-lite", "bytedance%2Fseed-2-0-lite"],
+  ["bytedance-seed%2Fdoubao-seed-code", "bytedance%2Fdoubao-seed-code"],
+  ["bytedance-seed%2Fseed-oss-36b-instruct", "ByteDance-Seed%2Fseed-oss-36b"],
+  ["claude-opus-4-7", "anthropic%2Fclaude-opus-4.7"],
+  ["codegen3_5k-qwen2.5-72b-instruct-2-chk-50", "codegen3-5k-qwen2.5-72b-instruct-2-chk-50"],
+  ["cognitivecomputations%2Fdolphin-2.9.2-qwen2-72b", "cognitivecomputations%2Fdolphin-2-9.2-qwen2-72b"],
+  ["cohere%2FAyaExpanse-32B", "CohereLabs%2Faya-expanse-32b"],
+  ["cohere%2FAyaExpanse-8B", "CohereLabs%2Faya-expanse-8b"],
+  ["cohere%2Faya-expanse-32b", "CohereLabs%2Faya-expanse-32b"],
+  ["cohere%2Faya-expanse-8b", "CohereLabs%2Faya-expanse-8b"],
+  ["cohere%2Fc4ai-command-r-v01", "CohereLabs%2Fc4ai-command-r-v01"],
+  ["cohere%2Fcommand-a-03-2025", "cohere%2Fcommand-a"],
+  ["cohere%2Fcommand-a-fc", "cohere%2Fcommand-a"],
+  ["cohere%2Fcommand-a-reasoning-fc", "cohere%2Fcommand-a-reasoning"],
+  ["cohere%2Fcommand-r-03-2024", "cohere%2Fcommand-r"],
+  ["cohere%2Fcommand-r-08-2024", "cohere%2Fcommand-r"],
+  ["cohere%2Fcommand-r-plus", "CohereLabs%2Fc4ai-command-r-plus"],
+  ["cohere%2Fcommand-r-plus-04-2024", "CohereLabs%2Fc4ai-command-r-plus"],
+  ["cohere%2Fcommand-r-plus-08-2024", "cohere%2Fcohere-command-r-plus-08-2024"],
+  ["cohere%2Fcommand-r7b-fc", "cohere%2Fcommand-r7b"],
+  ["cohere%2Ftiny-aya-global", "CohereLabs%2Ftiny-aya-global"],
+  ["databricks%2Fdbrx-base", "databricks%2Fdbrx"],
+  ["databricks%2Fdbrx-instruct", "databricks%2Fdbrx"],
+  ["deepcogito%2Fcogito-v2-1-reasoning", "deepcogito%2Fcogito-v2.1"],
+  ["deepseek%2FDeepSeek-R1", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2FDeepSeek-R1-0528", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2FDeepSeek-V3", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek%2FDeepSeek-V3-0324", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek%2Fchat-v3-0324", "deepseek%2Fdeepseek-chat-v3-0324"],
+  ["deepseek%2Fdeepseek-chat", "deepseek%2Fdeepseek"],
+  ["deepseek%2Fdeepseek-coder-v2-instruct", "deepseek%2Fdeepseek-coder-v2"],
+  ["deepseek%2Fdeepseek-coder-v2-lite-instruct", "deepseek%2Fdeepseek-coder-v2-lite"],
+  ["deepseek%2Fdeepseek-llm-67b-chat", "deepseek%2Fdeepseek-llm-67b"],
+  ["deepseek%2Fdeepseek-r1", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2Fdeepseek-r1-0120", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2Fdeepseek-r1-0528", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2Fdeepseek-r1-distill-llama-70b", "deepseek-ai%2FDeepSeek-R1-Distill-Llama-70B"],
+  ["deepseek%2Fdeepseek-r1-distill-llama-8b", "deepseek-ai%2FDeepSeek-R1-Distill-Llama-8B"],
+  ["deepseek%2Fdeepseek-r1-distill-qwen-1-5b", "deepseek-ai%2FDeepSeek-R1-Distill-Qwen-1.5B"],
+  ["deepseek%2Fdeepseek-r1-distill-qwen-14b", "deepseek-ai%2FDeepSeek-R1-Distill-Qwen-14B"],
+  ["deepseek%2Fdeepseek-r1-distill-qwen-32b", "deepseek-ai%2FDeepSeek-R1-Distill-Qwen-32B"],
+  ["deepseek%2Fdeepseek-v2-chat", "deepseek%2Fdeepseek-v2"],
+  ["deepseek%2Fdeepseek-v3", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek%2Fdeepseek-v3-0324", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek%2Fdeepseek-v3-1-reasoning", "deepseek%2Fdeepseek-v3-1"],
+  ["deepseek%2Fdeepseek-v3-1-terminus", "deepseek-ai%2FDeepSeek-V3.1-Terminus"],
+  ["deepseek%2Fdeepseek-v3-2", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek%2Fdeepseek-v3-2-0925", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek%2Fdeepseek-v3-2-exp-fc", "deepseek%2Fdeepseek-v3-2-exp"],
+  ["deepseek%2Fdeepseek-v3-2-reasoning", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek%2Fdeepseek-v3-2-reasoning-0925", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek%2Fdeepseek-v3-2-speciale", "deepseek-ai%2FDeepSeek-V3.2-Speciale"],
+  ["deepseek%2Fdeepseek-v3.1", "deepseek%2Fdeepseek-v3-1"],
+  ["deepseek%2Fdeepseek-v3.2", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek%2Fdeepseek-v3.2-exp", "deepseek%2Fdeepseek-v3-2-exp"],
+  ["deepseek%2Fdeepseek-v3.2-speciale", "deepseek-ai%2FDeepSeek-V3.2-Speciale"],
+  ["deepseek%2Fdeepseek-v4-flash-fp8", "deepseek%2Fdeepseek-v4-flash"],
+  ["deepseek%2Fdeepseek-v4-pro", "deepseek-ai%2FDeepSeek-V4-Pro"],
+  ["deepseek%2Fr1", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2Fr1-0528-openrouter", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek%2Fv3.2", "deepseek-ai%2FDeepSeek-V3.2"],
+  ["deepseek-ai%2FDeepSeek-R1-0528", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek-ai%2Fdeepseek-llm-67b-chat", "deepseek%2Fdeepseek-llm-67b"],
+  ["deepseek-ai%2Fdeepseek-llm-7b-chat", "deepseek%2Fdeepseek-llm-7b"],
+  ["deepseek-ai%2Fdeepseek-r1", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek-ai%2Fdeepseek-r1-0528", "deepseek-ai%2FDeepSeek-R1"],
+  ["deepseek-ai%2Fdeepseek-v3", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek-chat", "deepseek%2Fdeepseek"],
+  ["deepseek-v3", "deepseek-ai%2FDeepSeek-V3"],
+  ["deepseek-v3.1", "deepseek%2Fdeepseek-v3-1"],
+  ["dwikitheduck%2Fgemma-2-2b-id", "dwikitheduck%2Fgemma-2-2b-id-inst"],
+  ["dwikitheduck%2Fgemma-2-2b-id-instruct", "dwikitheduck%2Fgemma-2-2b-id-inst"],
+  ["eleutherai%2FPythia-12B", "EleutherAI%2Fpythia-12b"],
+  ["eleutherai%2FPythia-6.9B", "EleutherAI%2Fpythia-6.9b"],
+  ["eleutherai%2Fpythia-12b-v0", "EleutherAI%2Fpythia-12b"],
+  ["eleutherai%2Fpythia-6.9b", "EleutherAI%2Fpythia-6.9b"],
+  ["ewre324%2FThinker-Llama-3.2-3B-Instruct-Reasoning", "ewre324%2Fewre324-Thinker-Llama-3.2-3B-Instruct-Reasoning"],
+  ["ewre324%2FThinker-Qwen2.5-0.5B-Instruct-Reasoning", "ewre324%2Fewre324-Thinker-Qwen2.5-0.5B-Instruct-Reasoning"],
+  ["ewre324%2FThinker-SmolLM2-135M-Instruct-Reasoning", "ewre324%2Fewre324-Thinker-SmolLM2-135M-Instruct-Reasoning"],
+  ["formulae%2Fmita-v1-7b", "formulae%2Fmita-v1.0-7b-2-24-2025"],
+  ["gemini-2.5-pro-exp-03-25", "google%2Fgemini-2-5-pro-exp-03-25"],
+  ["gemma-1.1-7b-it", "google%2Fgemma-1.1-7b"],
+  ["google%2FGemini%202.5%20Flash", "google%2Fgemini-2.5-flash"],
+  ["google%2FGemini%202.5%20Pro", "google%2Fgemini-2.5-pro"],
+  ["google%2FGemini%203%20Flash", "google%2Fgemini-3-flash"],
+  ["google%2FGemini%203%20Pro", "google%2Fgemini-3-pro"],
+  ["google%2FGemini%203.1%20Pro", "google%2Fgemini-3.1-pro"],
+  ["google%2FGemini-2.5-Pro", "google%2Fgemini-2.5-pro"],
+  ["google%2FGemma-3-12B", "google%2Fgemma3-12b"],
+  ["google%2FGemma-3-27B", "google%2Fgemma-3-27b"],
+  ["google%2Fflan-t5-large", "google%2Ft5-large"],
+  ["google%2Fgemini-1.0-pro-001", "google%2Fgemini-1.0-pro"],
+  ["google%2Fgemini-1.0-pro-002", "google%2Fgemini-1.0-pro"],
+  ["google%2Fgemini-1.5-flash-001", "google%2Fgemini-1.5-flash"],
+  ["google%2Fgemini-1.5-flash-002", "google%2Fgemini-1.5-flash"],
+  ["google%2Fgemini-1.5-pro-001", "google%2Fgemini-1.5-pro"],
+  ["google%2Fgemini-1.5-pro-002", "google%2Fgemini-1.5-pro"],
+  ["google%2Fgemini-2-0-flash", "google%2Fgemini-2.0-flash"],
+  ["google%2Fgemini-2-0-flash-lite-001", "google%2Fgemini-2.0-flash-lite"],
+  ["google%2Fgemini-2-5-flash", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2-5-flash-fc", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2-5-flash-lite", "google%2Fgemini-2.5-flash-lite"],
+  ["google%2Fgemini-2-5-flash-lite-fc", "google%2Fgemini-2.5-flash-lite"],
+  ["google%2Fgemini-2-5-flash-lite-prompt", "google%2Fgemini-2.5-flash-lite"],
+  ["google%2Fgemini-2-5-flash-lite-reasoning", "google%2Fgemini-2.5-flash-lite"],
+  ["google%2Fgemini-2-5-flash-prompt", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2-5-flash-reasoning", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2-5-flash-reasoning-04-2025", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2-5-pro", "google%2Fgemini-2.5-pro"],
+  ["google%2Fgemini-2.0-flash-001", "google%2Fgemini-2.0-flash"],
+  ["google%2Fgemini-2.0-flash-lite-001", "google%2Fgemini-2.0-flash-lite"],
+  ["google%2Fgemini-2.0-flash-thinking", "google%2Fgemini-2.0-flash"],
+  ["google%2Fgemini-2.0-flash-thinking-exp-01-21", "google%2Fgemini-2-0-flash-thinking-exp-01-21"],
+  ["google%2Fgemini-2.0-flash-thinking-exp-1219", "google%2Fgemini-2-0-flash-thinking-exp-1219"],
+  ["google%2Fgemini-2.0-pro", "google%2Fgemini-2-0-pro"],
+  ["google%2Fgemini-2.0-pro-exp-02-05", "google%2Fgemini-2-0-pro-exp-02-05"],
+  ["google%2Fgemini-2.5-flash-lite-preview-09-2025", "google%2Fgemini-2-5-flash-lite-preview-09-2025"],
+  ["google%2Fgemini-2.5-flash-preview-04-17", "google%2Fgemini-2-5-flash-preview-04-17"],
+  ["google%2Fgemini-2.5-flash-preview-09-2025", "google%2Fgemini-2-5-flash-preview-09-2025"],
+  ["google%2Fgemini-2.5-flash-thinking", "google%2Fgemini-2.5-flash"],
+  ["google%2Fgemini-2.5-pro-exp-03-25", "google%2Fgemini-2-5-pro-exp-03-25"],
+  ["google%2Fgemini-2.5-pro-preview", "google%2Fgemini-2-5-pro-preview"],
+  ["google%2Fgemini-3-flash-preview", "google%2Fgemini-3-flash"],
+  ["google%2Fgemini-3-pro-low", "google%2Fgemini-3-pro"],
+  ["google%2Fgemini-3-pro-preview", "google%2Fgemini-3-pro"],
+  ["google%2Fgemini-3-pro-preview-fc", "google%2Fgemini-3-pro"],
+  ["google%2Fgemini-3-pro-preview-prompt", "google%2Fgemini-3-pro"],
+  ["google%2Fgemini-exp-1206", "google%2Fgemini-exp"],
+  ["google%2Fgemma-1.1-2b-it", "google%2Fgemma-1.1-2b"],
+  ["google%2Fgemma-1.1-7b-it", "google%2Fgemma-1.1-7b"],
+  ["google%2Fgemma-2-27b-it", "google%2Fgemma-2-27b"],
+  ["google%2Fgemma-2-2b-it", "google%2Fgemma-2-2b"],
+  ["google%2Fgemma-2-9b-it", "google%2Fgemma-2-9b"],
+  ["google%2Fgemma-2b-it", "google%2Fgemma-2b"],
+  ["google%2Fgemma-3-12b", "google%2Fgemma3-12b"],
+  ["google%2Fgemma-3-12b-it", "google%2Fgemma3-12b"],
+  ["google%2Fgemma-3-12b-it-prompt", "google%2Fgemma3-12b"],
+  ["google%2Fgemma-3-1b-it", "google%2Fgemma-3-1b"],
+  ["google%2Fgemma-3-1b-it-prompt", "google%2Fgemma-3-1b"],
+  ["google%2Fgemma-3-27b-it", "google%2Fgemma-3-27b"],
+  ["google%2Fgemma-3-27b-it-prompt", "google%2Fgemma-3-27b"],
+  ["google%2Fgemma-3-27b-nothink", "google%2Fgemma-3-27b"],
+  ["google%2Fgemma-3-4b", "google%2Fgemma3-4b"],
+  ["google%2Fgemma-3-4b-it", "google%2Fgemma3-4b"],
+  ["google%2Fgemma-3-4b-it-prompt", "google%2Fgemma3-4b"],
+  ["google%2Fgemma-3n-e2b", "google%2Fgemma-3n-E2B"],
+  ["google%2Fgemma-3n-e2b-it", "google%2Fgemma-3n-E2B"],
+  ["google%2Fgemma-3n-e2b-it-litert-preview", "google%2Fgemma-3n-E2B"],
+  ["google%2Fgemma-3n-e4b", "google%2Fgemma-3n-E4B"],
+  ["google%2Fgemma-3n-e4b-it", "google%2Fgemma-3n-E4B"],
+  ["google%2Fgemma-3n-e4b-it-litert-preview", "google%2Fgemma-3n-E4B"],
+  ["google%2Fgemma-4-26B-A4B-it", "google%2Fgemma4-26b"],
+  ["google%2Fgemma-4-26b-a4b", "google%2Fgemma4-26b"],
+  ["google%2Fgemma-4-26b-a4b-it", "google%2Fgemma4-26b"],
+  ["google%2Fgemma-4-26b-a4b-non-reasoning", "google%2Fgemma4-26b"],
+  ["google%2Fgemma-4-31B-it", "google%2Fgemma-4-31B"],
+  ["google%2Fgemma-4-31b", "google%2Fgemma-4-31B"],
+  ["google%2Fgemma-4-31b-it", "google%2Fgemma-4-31B"],
+  ["google%2Fgemma-4-31b-non-reasoning", "google%2Fgemma-4-31B"],
+  ["google%2Fgemma-4-E2B-it", "google%2Fgemma-4-E2B"],
+  ["google%2Fgemma-4-E4B-it", "google%2Fgemma-4-E4B"],
+  ["google%2Fgemma-4-e2b", "google%2Fgemma-4-E2B"],
+  ["google%2Fgemma-4-e2b-it", "google%2Fgemma-4-E2B"],
+  ["google%2Fgemma-4-e2b-non-reasoning", "google%2Fgemma-4-E2B"],
+  ["google%2Fgemma-4-e4b", "google%2Fgemma-4-E4B"],
+  ["google%2Fgemma-4-e4b-it", "google%2Fgemma-4-E4B"],
+  ["google%2Fgemma-4-e4b-non-reasoning", "google%2Fgemma-4-E4B"],
+  ["google%2Fgemma-7b-it", "google%2Fgemma-7b"],
+  ["google%2Fgemma4-31b", "google%2Fgemma-4-31B"],
+  ["google%2Fgemma4-e2b", "google%2Fgemma-4-E2B"],
+  ["google%2Fgemma4-e4b", "google%2Fgemma-4-E4B"],
+  ["google%2Flearnlm-1.5-pro-experimental", "google%2Flearnlm-1-5-pro-experimental"],
+  ["google%2Fmedgemma-4b-it", "google%2Fmedgemma-4b"],
+  ["google%2Ftext-bison%40001", "google%2Ftext-bison"],
+  ["google%2Ftext-unicorn%40001", "google%2Ftext-unicorn"],
+  ["gpt-4o", "openai%2Fgpt-4o"],
+  ["gpt-4o-mini", "openai%2Fgpt-4o-mini"],
+  ["gpt-5", "openai%2Fgpt-5"],
+  ["grok-3", "xai%2Fgrok-3"],
+  ["hermes-3-llama-3.1-70b", "NousResearch%2FHermes-3-Llama-3.1-70B"],
+  ["huggingface%2Fsmollm2-1.7b", "HuggingFaceTB%2FSmolLM2-1.7B"],
+  ["huggingface%2Fsmollm2-135m", "HuggingFaceTB%2FSmolLM2-135M"],
+  ["huggingface%2Fsmollm3-3b-nothink", "huggingface%2Fsmollm3-3b"],
+  ["iRyanBell%2FARC1", "iRyanBell%2Fllama3.0-ARC1-8b"],
+  ["iRyanBell%2FARC1-II", "iRyanBell%2Fllama3.0-ARC1-II-8b"],
+  ["ibm%2FPowerLM-3b", "ibm-research%2FPowerLM-3b"],
+  ["ibm%2Fgranite-3.0-2b-base", "ibm-granite%2Fgranite-3.0-2b-base"],
+  ["ibm%2Fgranite-3.0-8b-base", "ibm-granite%2Fgranite-3.0-8b-base"],
+  ["ibm%2Fgranite-3.1-1b-a400m-base", "ibm-granite%2Fgranite-3.1-1b-a400m-base"],
+  ["ibm%2Fgranite-3.1-1b-a400m-instruct", "ibm-granite%2Fgranite-3.1-1b-a400m-instruct"],
+  ["ibm%2Fgranite-3.1-2b-base", "ibm-granite%2Fgranite-3.1-2b-base"],
+  ["ibm%2Fgranite-3.1-2b-instruct", "ibm-granite%2Fgranite-3.1-2b-instruct"],
+  ["ibm%2Fgranite-3.1-3b-a800m-base", "ibm-granite%2Fgranite-3.1-3b-a800m-base"],
+  ["ibm%2Fgranite-3.1-3b-a800m-instruct", "ibm-granite%2Fgranite-3.1-3b-a800m-instruct"],
+  ["ibm%2Fgranite-3.1-8b-base", "ibm-granite%2Fgranite-3.1-8b-base"],
+  ["ibm%2Fgranite-3.1-8b-instruct", "ibm-granite%2Fgranite-3.1-8b"],
+  ["ibm%2Fgranite-3.3-8b-base", "ibm-granite%2Fgranite-3.3-8b"],
+  ["ibm%2Fgranite-3.3-8b-instruct", "ibm-granite%2Fgranite-3.3-8b"],
+  ["ibm%2Fgranite-4-0-micro", "ibm%2Fgranite-4-0-h-micro"],
+  ["ibm%2Fgranite-4.0-micro", "ibm%2Fgranite-4-0-h-micro"],
+  ["ibm%2Fmerlinite-7b", "ibm-research%2Fmerlinite-7b"],
+  ["ibm-granite%2Fgranite-3.0-2b-instruct", "ibm-granite%2Fgranite-3.0-2b"],
+  ["ibm-granite%2Fgranite-3.0-8b-instruct", "ibm-granite%2Fgranite-3.0-8b"],
+  ["ibm-granite%2Fgranite-3.1-8b-instruct", "ibm-granite%2Fgranite-3.1-8b"],
+  ["ibm-granite%2Fgranite-3.2-8b-instruct", "ibm-granite%2Fgranite-3.2-8b"],
+  ["inclusionai%2Fling-1t", "inclusionAI%2FLing-1T"],
+  ["inclusionai%2Fling-flash-2-0", "inclusionAI%2Fling-flash-2-0"],
+  ["inclusionai%2Fling-mini-2-0", "inclusionAI%2Fling-mini-2-0"],
+  ["inclusionai%2Fring-1t", "inclusionAI%2FRing-1T"],
+  ["inclusionai%2Fring-flash-2-0", "inclusionAI%2Fring-flash-2-0"],
+  ["internlm%2Finternlm2_5-20b-chat", "internlm%2Finternlm2.5-20b"],
+  ["jeffmeloy%2Fjeffmeloy_Qwen2.5-7B-minperplexity-1", "jeffmeloy%2FQwen2.5-7B-minperplexity-1"],
+  ["katanemo%2Farch-agent-32b", "katanemo%2FArch-Agent-32B"],
+  ["katanemo%2Farch-agent-3b", "katanemo%2FArch-Agent-3B"],
+  ["kimi%2Fkimi-k2-5", "moonshotai%2FKimi-K2.5"],
+  ["kimi%2Fkimi-k2-thinking", "moonshotai%2FKimi-K2-Thinking"],
+  ["kwaikat%2Fkat-coder-pro-v1", "kwaikat%2Fkat-coder-pro"],
+  ["kwaikat%2Fkat-coder-pro-v2", "kwaikat%2Fkat-coder-pro"],
+  ["lg%2Fexaone-4-0-1-2b-reasoning", "lg%2Fexaone-4-0-1-2b"],
+  ["lg%2Fexaone-4-0-32b-reasoning", "lg%2Fexaone-4-0-32b"],
+  ["lg%2Fk-exaone-non-reasoning", "lg%2Fk-exaone"],
+  ["liquidai%2Flfm2-24b-a2b", "LiquidAI%2FLFM2-24B-A2B"],
+  ["liquidai%2Flfm2-8b-a1b", "LiquidAI%2FLFM2-8B-A1B"],
+  ["llama-2-7b-chat-hf", "meta-llama%2FLlama-2-7b-chat-hf"],
+  ["llama-3.3-70b-instruct", "meta-llama%2Fllama-3.3"],
+  ["llama-4-maverick-17b-128e-instruct", "meta-llama%2Fllama-4"],
+  ["lunahr%2Fthea-v2-3b-50r", "lunahr%2Funavailablemodel-thea-v2-3b-50r"],
+  ["mathstral-7B-v0.1", "mistralai%2FMathstral-7B-v0.1"],
+  ["mattshumer%2FReflection-70B", "mattshumer%2FReflection-Llama-3.1-70B"],
+  ["meta%2FLlama-2-13B", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-2-13b-chat", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-2-70B", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-2-70b-chat", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-2-7B", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-2-7b-chat", "meta-llama%2Fllama-2"],
+  ["meta%2FLlama-3.1-8B", "meta-llama%2Fllama-3.1"],
+  ["meta%2FLlama-4-Maverick", "meta%2Fllama-4-maverick"],
+  ["meta%2Fllama-13b", "meta%2FLLaMA-13B"],
+  ["meta%2Fllama-2-13b", "meta-llama%2Fllama-2"],
+  ["meta%2Fllama-2-13b-hf", "meta-llama%2FLlama-2-13b-hf"],
+  ["meta%2Fllama-2-70b", "meta-llama%2Fllama-2"],
+  ["meta%2Fllama-2-70b-hf", "meta-llama%2FLlama-2-70b-hf"],
+  ["meta%2Fllama-2-7b", "meta-llama%2Fllama-2"],
+  ["meta%2Fllama-2-7b-hf", "meta-llama%2FLlama-2-7b-hf"],
+  ["meta%2Fllama-3-1-8b-instruct-prompt", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3-2-1b-instruct-fc", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3-2-3b-instruct-fc", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3-3-70b-instruct-fc", "meta-llama%2Fllama-3.3"],
+  ["meta%2Fllama-3-70b", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3-70b-chat", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3-70b-instruct", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3-8b", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3-8b-chat", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3-8b-instruct", "meta-llama%2Fllama-3"],
+  ["meta%2Fllama-3.1-405b", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-405b-instruct", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-405b-instruct-turbo", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-70b", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-70b-instruct", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-70b-instruct-turbo", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-8b", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-8b-instruct", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.1-8b-instruct-turbo", "meta-llama%2Fllama-3.1"],
+  ["meta%2Fllama-3.2-11b-instruct", "meta%2Fllama-3-2-11b"],
+  ["meta%2Fllama-3.2-11b-vision-instruct-turbo", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.2-1b", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.2-3b", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.2-3b-instruct", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.2-90b-instruct", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.2-90b-vision-instruct-turbo", "meta-llama%2Fllama-3.2"],
+  ["meta%2Fllama-3.3-70b-instruct", "meta-llama%2Fllama-3.3"],
+  ["meta%2Fllama-3.3-70b-instruct-turbo", "meta-llama%2Fllama-3.3"],
+  ["meta%2Fllama-30b", "meta%2FLLaMA-30B"],
+  ["meta%2Fllama-4-maverick-17b-128e-instruct-fp8", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-4-maverick-17b-128e-instruct-fp8-fc", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-4-maverick-17b-128e-instruct-fp8-together", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-4-scout-17b-16e-instruct", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-4-scout-17b-16e-instruct-fc", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-4-scout-17b-16e-instruct-together", "meta-llama%2Fllama-4"],
+  ["meta%2Fllama-65b", "meta%2FLLaMA-65B"],
+  ["meta%2Fllama-7b", "meta%2FLLaMA-7B"],
+  ["meta%2Fllama3.1-8b", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FLlama-3.1-70B", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FLlama-3.1-70B-Instruct", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FLlama-3.1-8B", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FLlama-3.1-8B-Instruct", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FLlama-3.2-1B", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2FLlama-3.2-1B-Instruct", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2FLlama-3.2-3B", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2FLlama-3.2-3B-Instruct", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2FLlama-3.2-90B-Vision-Instruct", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2FLlama-3.3-70B-Instruct", "meta-llama%2Fllama-3.3"],
+  ["meta-llama%2FLlama-4-Maverick-17B-128E-Instruct-FP8", "meta-llama%2Fllama-4"],
+  ["meta-llama%2FLlama-4-Scout-17B-16E-Instruct", "meta-llama%2Fllama-4"],
+  ["meta-llama%2FMeta-Llama-3-70B", "meta-llama%2Fllama-3"],
+  ["meta-llama%2FMeta-Llama-3-70B-Instruct", "meta-llama%2Fllama-3"],
+  ["meta-llama%2FMeta-Llama-3-8B", "meta-llama%2Fllama-3"],
+  ["meta-llama%2FMeta-Llama-3-8B-Instruct", "meta-llama%2Fllama-3"],
+  ["meta-llama%2FMeta-Llama-3.1-70B-Instruct", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FMeta-Llama-3.1-8B", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2FMeta-Llama-3.1-8B-Instruct", "meta-llama%2Fllama-3.1"],
+  ["meta-llama%2Fllama-3.2-90b-vision-instruct", "meta-llama%2Fllama-3.2"],
+  ["meta-llama%2Fllama-3.3-70b-instruct", "meta-llama%2Fllama-3.3"],
+  ["meta-llama%2Fllama3.1-8b", "meta-llama%2Fllama-3.1"],
+  ["meta-llama-3-70b-instruct", "meta-llama%2Fllama-3"],
+  ["meta-llama-3-8b-instruct", "meta-llama%2Fllama-3"],
+  ["microsoft%2FPhi-3-medium-128k-instruct", "microsoft%2Fphi-3-medium-128k"],
+  ["microsoft%2FPhi-3-medium-4k-instruct", "microsoft%2Fphi-3-medium-4k"],
+  ["microsoft%2FPhi-3-mini-128k-instruct", "microsoft%2Fphi-3-mini-128k"],
+  ["microsoft%2FPhi-3-mini-4k-instruct", "microsoft%2Fphi-3-mini-4k"],
+  ["microsoft%2FPhi-3-small-128k-instruct", "microsoft%2Fphi-3-small-128k"],
+  ["microsoft%2FPhi-3-small-8k-instruct", "microsoft%2Fphi-3-small-8k"],
+  ["microsoft%2FPhi-3.5-MoE-instruct", "microsoft%2Fphi-3.5-moe"],
+  ["microsoft%2FPhi-3.5-mini-instruct", "microsoft%2Fphi-3.5-mini"],
+  ["microsoft%2FPhi-4", "microsoft%2Fphi-4"],
+  ["microsoft%2FPhi-4-mini-instruct", "microsoft%2FPhi-4-mini-reasoning"],
+  ["microsoft%2FPhi-4-reasoning", "microsoft%2Fphi-4"],
+  ["microsoft%2Fphi-3-medium-4k-instruct", "microsoft%2Fphi-3-medium-4k"],
+  ["microsoft%2Fphi-3-small-8k-instruct", "microsoft%2Fphi-3-small-8k"],
+  ["microsoft%2Fphi-3.5-mini-instruct", "microsoft%2Fphi-3.5-mini"],
+  ["microsoft%2Fphi-3.5-moe-instruct", "microsoft%2Fphi-3.5-moe"],
+  ["microsoft%2Fphi-3.5-vision-instruct", "microsoft%2Fphi-3.5-vision"],
+  ["microsoft%2Fphi-4-mini", "microsoft%2FPhi-4-mini-reasoning"],
+  ["microsoft%2Fphi-4-mini-reasoning", "microsoft%2FPhi-4-mini-reasoning"],
+  ["microsoft%2Fphi-4-multimodal-instruct", "microsoft%2Fphi-4-multimodal"],
+  ["microsoft%2Fphi-4-prompt", "microsoft%2Fphi-4"],
+  ["microsoft%2Fphi-4-reasoning", "microsoft%2Fphi-4"],
+  ["microsoft%2Fphi-4-reasoning-plus", "microsoft%2Fphi-4"],
+  ["microsoft%2Fphi3-medium-128k", "microsoft%2Fphi-3-medium-128k"],
+  ["microsoft%2Fphi3-medium-4k", "microsoft%2Fphi-3-medium-4k"],
+  ["microsoft%2Fphi3-mini-128k", "microsoft%2Fphi-3-mini-128k"],
+  ["microsoft%2Fphi3-mini-4k", "microsoft%2Fphi-3-mini-4k"],
+  ["microsoft%2Fphi4-reasoning-plus", "microsoft%2Fphi-4"],
+  ["minimax%2FMiniMax-M2.1", "MiniMaxAI%2FMiniMax-M2.1"],
+  ["minimax%2FMiniMax-M2.5", "MiniMaxAI%2FMiniMax-M2.5"],
+  ["minimax%2FMiniMax-M2.5-Lightning", "minimax%2Fminimax-m2-5-lightning"],
+  ["minimax%2FMiniMax-M2.7", "minimax%2Fminimax-m2.7"],
+  ["minimax%2Fm2.5", "MiniMaxAI%2FMiniMax-M2.5"],
+  ["minimax%2Fminimax-m2-1", "MiniMaxAI%2FMiniMax-M2.1"],
+  ["minimax%2Fminimax-m2-5", "MiniMaxAI%2FMiniMax-M2.5"],
+  ["minimax%2Fminimax-m2-7", "minimax%2Fminimax-m2.7"],
+  ["minimax%2Fminimax-m2.1", "MiniMaxAI%2FMiniMax-M2.1"],
+  ["minimax%2Fminimax-m2.5", "MiniMaxAI%2FMiniMax-M2.5"],
+  ["mistral%2Fdevstral-2", "mistralai%2Fdevstral-2"],
+  ["mistral%2Fdevstral-medium", "mistralai%2Fdevstral-medium"],
+  ["mistral%2Fdevstral-small", "mistralai%2Fdevstral-small"],
+  ["mistral%2Fdevstral-small-2", "mistralai%2Fdevstral-small-2"],
+  ["mistral%2Fdevstral-small-2505", "mistralai%2Fdevstral-small"],
+  ["mistral%2Fmagistral-medium", "mistralai%2Fmagistral-medium"],
+  ["mistral%2Fmagistral-medium-2509", "mistralai%2Fmagistral-medium"],
+  ["mistral%2Fmagistral-small", "mistralai%2Fmagistral-small"],
+  ["mistral%2Fmagistral-small-2509", "mistralai%2Fmagistral-small"],
+  ["mistral%2Fministral-3-14b", "mistralai%2Fministral-3-14b"],
+  ["mistral%2Fministral-3-3b", "mistralai%2Fministral-3-3b"],
+  ["mistral%2Fministral-3-8b", "mistralai%2Fministral-3-8b"],
+  ["mistral%2Fministral-3-8b-base-2512", "mistralai%2Fministral-3-8b"],
+  ["mistral%2Fministral-3-8b-instruct-2512", "mistralai%2Fministral-3-8b"],
+  ["mistral%2Fministral-8b-instruct-2410", "mistralai%2Fministral-8b"],
+  ["mistral%2Fministral-8b-latest", "mistralai%2Fministral-8b"],
+  ["mistral%2Fmistral-7b-instruct", "mistralai%2Fmistral-7b"],
+  ["mistral%2Fmistral-large", "mistralai%2Fmistral-large"],
+  ["mistral%2Fmistral-large-2402", "mistralai%2Fmistral-large"],
+  ["mistral%2Fmistral-large-2407", "mistralai%2Fmistral-large"],
+  ["mistral%2Fmistral-large-3", "mistralai%2Fmistral-large-3"],
+  ["mistral%2Fmistral-large-3-2509", "mistralai%2Fmistral-large-3"],
+  ["mistral%2Fmistral-large-latest", "mistralai%2Fmistral-large"],
+  ["mistral%2Fmistral-medium", "mistralai%2Fmistral-medium"],
+  ["mistral%2Fmistral-medium-3", "mistralai%2Fmistral-medium-3"],
+  ["mistral%2Fmistral-medium-3-1", "mistralai%2Fmistral-medium-3.1"],
+  ["mistral%2Fmistral-nemo-instruct-2407", "mistralai%2Fmistral-nemo"],
+  ["mistral%2Fmistral-saba", "mistralai%2Fmistral-saba"],
+  ["mistral%2Fmistral-small", "mistralai%2Fmistral-small"],
+  ["mistral%2Fmistral-small-24b-base-2501", "mistralai%2FMistral-Small-24B-Base-2501"],
+  ["mistral%2Fmistral-small-24b-instruct-2501", "mistralai%2Fmistral-small-24b-instruct-2501"],
+  ["mistral%2Fmistral-small-2503", "mistralai%2Fmistral-small-2503"],
+  ["mistral%2Fmistral-small-3", "mistralai%2Fmistral-small-24b-instruct-2501"],
+  ["mistral%2Fmistral-small-3-1", "mistralai%2Fmistral-small-24b-instruct-2501"],
+  ["mistral%2Fmistral-small-3-2", "mistralai%2Fmistral-small-3.2"],
+  ["mistral%2Fmistral-small-3.2-24b-instruct-2506", "mistralai%2Fmistral-small-3-2-24b-instruct-2506"],
+  ["mistral%2Fmistral-small-4", "mistralai%2Fmistral-small-4"],
+  ["mistral%2Fmistral-small-4-non-reasoning", "mistralai%2Fmistral-small-4"],
+  ["mistral%2Fmistral-small-latest", "mistralai%2Fmistral-small"],
+  ["mistral%2Fmixtral-8x7b-instruct", "mistralai%2Fmixtral-8x7b"],
+  ["mistral%2Fpixtral-12b-2409", "mistralai%2Fpixtral-12b"],
+  ["mistral%2Fpixtral-large", "mistralai%2Fpixtral-large"],
+  ["mistral%2Fpixtral-large-2411", "mistralai%2Fpixtral-large"],
+  ["mistral-7b-instruct-v0.2", "mistralai%2FMistral-7B-Instruct-v0.2"],
+  ["mistralai%2FDevstral-Small-2505", "mistralai%2Fdevstral-small"],
+  ["mistralai%2FMinistral-8B-Instruct-2410", "mistralai%2Fministral-8b"],
+  ["mistralai%2FMistral-7B", "mistralai%2Fmistral-7b"],
+  ["mistralai%2FMistral-7B-Instruct-v0.3", "mistralai%2Fmistral-7b"],
+  ["mistralai%2FMistral-Medium", "mistralai%2Fmistral-medium"],
+  ["mistralai%2FMistral-Nemo-Base-2407", "mistralai%2Fmistral-nemo"],
+  ["mistralai%2FMistral-Nemo-Instruct-2407", "mistralai%2Fmistral-nemo"],
+  ["mistralai%2Fdevstral-small-2507", "mistralai%2Fdevstral-small"],
+  ["mistralai%2Fmagistral-small-2506", "mistralai%2FMagistral-Small-2506"],
+  ["mistralai%2Fmagistral-small-2509", "mistralai%2Fmagistral-small"],
+  ["mistralai%2Fministral-8b-instruct-2410-fc", "mistralai%2Fministral-8b"],
+  ["mistralai%2Fmistral-7b-instruct-v0.1", "mistralai%2FMistral-7B-Instruct-v0.1"],
+  ["mistralai%2Fmistral-7b-instruct-v0.2", "mistralai%2FMistral-7B-Instruct-v0.2"],
+  ["mistralai%2Fmistral-7b-instruct-v0.3", "mistralai%2Fmistral-7b"],
+  ["mistralai%2Fmistral-7b-v0.1", "mistralai%2FMistral-7B-v0.1"],
+  ["mistralai%2Fmistral-large-2407", "mistralai%2Fmistral-large"],
+  ["mistralai%2Fmistral-large-2411", "mistralai%2Fmistral-large"],
+  ["mistralai%2Fmistral-large-instruct-2407", "mistralai%2FMistral-Large-Instruct-2407"],
+  ["mistralai%2Fmistral-large-instruct-2411", "mistralai%2FMistral-Large-Instruct-2411"],
+  ["mistralai%2Fmistral-medium-2505", "mistralai%2Fmistral-medium"],
+  ["mistralai%2Fmistral-nemo-base-2407", "mistralai%2Fmistral-nemo"],
+  ["mistralai%2Fmistral-nemo-instruct-2407", "mistralai%2Fmistral-nemo"],
+  ["mistralai%2Fmistral-small-instruct", "mistralai%2Fmistral-small"],
+  ["mistralai%2Fmistral-small-instruct-2409", "mistralai%2FMistral-Small-Instruct-2409"],
+  ["mistralai%2Fmixtral-8x22b-instruct-v0.1", "mistralai%2FMixtral-8x22B-Instruct-v0.1"],
+  ["mistralai%2Fmixtral-8x7b-instruct-v0.1", "mistralai%2FMixtral-8x7B-Instruct-v0.1"],
+  ["mistralai%2Fmixtral-8x7b-v0.1", "mistralai%2FMixtral-8x7B-v0.1"],
+  ["mixtral-8x22b-instruct-v0.1", "mistralai%2FMixtral-8x22B-Instruct-v0.1"],
+  ["mixtral-8x7b-instruct-v0.1", "mistralai%2FMixtral-8x7B-Instruct-v0.1"],
+  ["moonshot%2FKimi%20K2%20Thinking", "moonshotai%2FKimi-K2-Thinking"],
+  ["moonshot%2FKimi%20K2.5", "moonshotai%2FKimi-K2.5"],
+  ["moonshot%2Fkimi-k2-0711-preview", "moonshotai%2Fkimi-k2"],
+  ["moonshot%2Fkimi-k2.5", "moonshotai%2FKimi-K2.5"],
+  ["moonshot-ai%2Fkimi-k2-instruct", "moonshotai%2Fkimi-k2"],
+  ["moonshot-ai%2Fkimi-k2-thinking", "moonshotai%2FKimi-K2-Thinking"],
+  ["moonshot-ai%2Fkimi-k2.5", "moonshotai%2FKimi-K2.5"],
+  ["moonshotai%2FKimi-K2-Instruct", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-0905", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-0905-preview", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-base", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-instruct", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-instruct-0905", "moonshotai%2Fkimi-k2"],
+  ["moonshotai%2Fkimi-k2-thinking", "moonshotai%2FKimi-K2-Thinking"],
+  ["moonshotai%2Fkimi-k2.5", "moonshotai%2FKimi-K2.5"],
+  ["moonshotai%2Fkimi-k2.5-thinking", "moonshotai%2FKimi-K2.5"],
+  ["moonshotai%2Fkimi-k2.6", "moonshotai%2FKimi-K2.6"],
+  ["moonshotai%2Fkimi-k2.6-thinking", "moonshotai%2FKimi-K2.6"],
+  ["moonshotai%2Fmoonshotai-kimi-k2-instruct-fc", "moonshotai%2Fkimi-k2"],
+  ["mosaicml%2Fmpt-30b", "mosaicml%2FMPT-30B"],
+  ["mosaicml%2Fmpt-30b-instruct", "mosaicml%2FMPT-30B"],
+  ["mosaicml%2Fmpt-instruct-30b", "mosaicml%2FMPT-Instruct-30B"],
+  ["my_model%2F", "my-model"],
+  ["nexusflow%2Fathene-v2-chat", "Nexusflow%2FAthene-V2-Chat"],
+  ["nhyha%2Fmerge_Qwen2.5-7B-Instruct_20241023_0314", "nhyha%2FN3N_Qwen2.5-7B-Instruct_20241023_0314"],
+  ["nous-research%2Fdeephermes-3-mistral-24b-preview", "NousResearch%2FDeepHermes-3-Mistral-24B-Preview"],
+  ["nous-research%2Fhermes-3-llama-3-1-70b", "NousResearch%2FHermes-3-Llama-3.1-70B"],
+  ["nous-research%2Fhermes-4-llama-3-1-405b-reasoning", "nous-research%2Fhermes-4-llama-3-1-405b"],
+  ["nous-research%2Fhermes-4-llama-3-1-70b-reasoning", "nous-research%2Fhermes-4-llama-3-1-70b"],
+  ["nousresearch%2Fnous-hermes-13b", "NousResearch%2FNous-Hermes-13b"],
+  ["nvidia%2Fllama-3-1-nemotron-nano-4b-reasoning", "nvidia%2Fllama-3.1-nemotron-nano-4b"],
+  ["nvidia%2Fllama-3-1-nemotron-ultra-253b-v1-reasoning", "nvidia%2Fllama-3-1-nemotron-ultra-253b"],
+  ["nvidia%2Fllama-3-3-nemotron-super-49b-reasoning", "nvidia%2Fllama-3-3-nemotron-super-49b"],
+  ["nvidia%2Fllama-3.1-nemotron-70b-instruct", "nvidia%2Fllama-3.1-nemotron-70b"],
+  ["nvidia%2Fllama-3.1-nemotron-nano-8b-v1", "nvidia%2Fllama-3.1-nemotron-nano-8b"],
+  ["nvidia%2Fllama-3.1-nemotron-ultra-253b-v1", "nvidia%2Fllama-3-1-nemotron-ultra-253b"],
+  ["nvidia%2Fllama-3.3-nemotron-super-49b-v1", "nvidia%2FLlama-3_3-Nemotron-Super-49B-v1"],
+  ["nvidia%2Fnemotron-3-nano-30b-a3b", "nvidia%2Fnemotron-3-nano-30b"],
+  ["nvidia%2Fnemotron-3-super-120b-a12b", "nvidia%2FNVIDIA-Nemotron-3-Super-120B-A12B-BF16"],
+  ["nvidia%2Fnemotron-cascade-2-30b-a3b", "nvidia%2FNemotron-Cascade-2-30B-A3B"],
+  ["nvidia%2Fnvidia-nemotron-3-nano-30b-a3b", "nvidia%2Fnemotron-3-nano-30b"],
+  ["nvidia%2Fnvidia-nemotron-3-super-120b-a12b", "nvidia%2FNVIDIA-Nemotron-3-Super-120B-A12B-FP8"],
+  ["nvidia%2Fnvidia-nemotron-nano-12b-v2-vl", "nvidia%2Fnemotron-nano-12b-vl"],
+  ["nvidia%2Fnvidia-nemotron-nano-9b-v2", "nvidia%2FNVIDIA-Nemotron-Nano-9B-v2"],
+  ["olmo-2-1124-13b-instruct", "allenai%2FOLMo-2-1124-13B"],
+  ["oopere%2Fpruned20-llama-1b", "oopere%2FLlama-3.2-1B-pruned-20pct"],
+  ["oopere%2Fpruned40-llama-1b", "oopere%2FLlama-3.2-1B-pruned-40pct"],
+  ["oopere%2Fpruned40-llama-3.2-1B", "oopere%2FLlama-3.2-1B-pruned-40pct"],
+  ["oopere%2Fpruned60-llama-1b", "oopere%2Fpruned60-llama-3.2-1B"],
+  ["openai%2FGPT%204o", "openai%2Fgpt-4o"],
+  ["openai%2FGPT%205", "openai%2Fgpt-5"],
+  ["openai%2FGPT%205%20Codex", "openai%2Fgpt-5-codex"],
+  ["openai%2FGPT%205.1", "openai%2Fgpt-5.1"],
+  ["openai%2FGPT%205.1%20Codex", "openai%2Fgpt-5.1-codex"],
+  ["openai%2FGPT%205.2", "openai%2Fgpt-5.2"],
+  ["openai%2FGPT%205.2%20Codex", "openai%2Fgpt-5.2-codex"],
+  ["openai%2FGPT%205.2%20Pro", "openai%2Fgpt-5.2-pro"],
+  ["openai%2FGPT%205.3%20Codex", "openai%2Fgpt-5.3-codex"],
+  ["openai%2FGPT%20OSS%20120B", "openai%2Fgpt-oss-120b"],
+  ["openai%2FGPT-4.1", "openai%2Fgpt-4.1"],
+  ["openai%2FGPT-4o-1120", "openai%2Fgpt-4o"],
+  ["openai%2FGPT-5", "openai%2Fgpt-5"],
+  ["openai%2FGPT-5.2", "openai%2Fgpt-5.2"],
+  ["openai%2FGPT-J-6B", "EleutherAI%2Fgpt-j-6b"],
+  ["openai%2FGPT-NeoX-20B", "EleutherAI%2Fgpt-neox-20b"],
+  ["openai%2FOpenAI-o1", "openai%2Fo1"],
+  ["openai%2FOpenAI-o3-mini-high", "openai%2Fo3-mini"],
+  ["openai%2Fgpt-3-5-turbo-0613", "openai%2Fgpt-3.5-turbo"],
+  ["openai%2Fgpt-3.5-turbo-0125", "openai%2Fgpt-3.5-turbo"],
+  ["openai%2Fgpt-3.5-turbo-0301", "openai%2Fgpt-3.5-turbo"],
+  ["openai%2Fgpt-3.5-turbo-0613", "openai%2Fgpt-3.5-turbo"],
+  ["openai%2Fgpt-3.5-turbo-1106", "openai%2Fgpt-3.5-turbo"],
+  ["openai%2Fgpt-4-0314", "openai%2Fgpt-4"],
+  ["openai%2Fgpt-4-0613", "openai%2Fgpt-4"],
+  ["openai%2Fgpt-4-1", "openai%2Fgpt-4.1"],
+  ["openai%2Fgpt-4-1-mini", "openai%2Fgpt-4.1"],
+  ["openai%2Fgpt-4-1-nano", "openai%2Fgpt-4.1"],
+  ["openai%2Fgpt-4.1-mini", "openai%2Fgpt-4.1"],
+  ["openai%2Fgpt-4.1-mini-2025-04-14", "openai%2Fgpt-4.1"],
+  ["openai%2Fgpt-4o-2024-05-13", "openai%2Fgpt-4o"],
+  ["openai%2Fgpt-4o-2024-08-06", "openai%2Fgpt-4o"],
+  ["openai%2Fgpt-4o-2024-11-20", "openai%2Fgpt-4o"],
+  ["openai%2Fgpt-4o-mini-2024-07-18", "openai%2Fgpt-4o-mini"],
+  ["openai%2Fgpt-5-1", "openai%2Fgpt-5.1"],
+  ["openai%2Fgpt-5-1-codex", "openai%2Fgpt-5.1-codex"],
+  ["openai%2Fgpt-5-1-codex-mini", "openai%2Fgpt-5.1-codex-mini"],
+  ["openai%2Fgpt-5-1-non-reasoning", "openai%2Fgpt-5.1"],
+  ["openai%2Fgpt-5-2", "openai%2Fgpt-5.2"],
+  ["openai%2Fgpt-5-2-codex", "openai%2Fgpt-5.2-codex"],
+  ["openai%2Fgpt-5-2-medium", "openai%2Fgpt-5.2"],
+  ["openai%2Fgpt-5-3-codex", "openai%2Fgpt-5.3-codex"],
+  ["openai%2Fgpt-5-4", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5-4-high", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5-4-low", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5-4-medium", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5-4-mini", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5-4-mini-high", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5-4-mini-low", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5-4-mini-medium", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5-4-mini-non-reasoning", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5-4-nano", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-5-4-nano-high", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-5-4-nano-low", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-5-4-nano-medium", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-5-4-nano-non-reasoning", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-5-4-non-reasoning", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5-4-pro", "openai%2Fgpt-5.4-pro"],
+  ["openai%2Fgpt-5-high", "openai%2Fgpt-5"],
+  ["openai%2Fgpt-5-low", "openai%2Fgpt-5"],
+  ["openai%2Fgpt-5-medium", "openai%2Fgpt-5"],
+  ["openai%2Fgpt-5-mini-medium", "openai%2Fgpt-5-mini"],
+  ["openai%2Fgpt-5-mini-minimal", "openai%2Fgpt-5-mini"],
+  ["openai%2Fgpt-5-minimal", "openai%2Fgpt-5"],
+  ["openai%2Fgpt-5-nano-medium", "openai%2Fgpt-5-nano"],
+  ["openai%2Fgpt-5-nano-minimal", "openai%2Fgpt-5-nano"],
+  ["openai%2Fgpt-5.1-2025-11-13", "openai%2Fgpt-5.1"],
+  ["openai%2Fgpt-5.1-codex-high", "openai%2Fgpt-5.1-codex"],
+  ["openai%2Fgpt-5.1-instant", "openai%2Fgpt-5-1-instant"],
+  ["openai%2Fgpt-5.1-thinking", "openai%2Fgpt-5.1"],
+  ["openai%2Fgpt-5.4-high", "openai%2Fgpt-5.4"],
+  ["openai%2Fgpt-5.4-mini-2026-03-17", "openai%2Fgpt-5.4-mini"],
+  ["openai%2Fgpt-5.4-nano-2026-03-17", "openai%2Fgpt-5.4-nano"],
+  ["openai%2Fgpt-oss-120b-high", "openai%2Fgpt-oss-120b"],
+  ["openai%2Fgpt-oss-120b-low", "openai%2Fgpt-oss-120b"],
+  ["openai%2Fgpt-oss-20b-high", "openai%2Fgpt-oss-20b"],
+  ["openai%2Fgpt-oss-20b-low", "openai%2Fgpt-oss-20b"],
+  ["openai%2Fgpt-oss-20b-medium", "openai%2Fgpt-oss-20b"],
+  ["openai%2Fo1-preview", "openai%2Fo1"],
+  ["openai%2Fo3%20Pro", "openai%2Fo3-pro"],
+  ["openbmb%2Fminicpm-sala", "openbmb%2FMiniCPM-SALA"],
+  ["openchat%2Fopenchat-3.5-0106", "openchat%2Fopenchat_3.5"],
+  ["openchat%2Fopenchat-3.5-1210", "openchat%2Fopenchat_3.5"],
+  ["openchat%2Fopenchat-35", "openchat%2Fopenchat_3.5"],
+  ["perplexity-sonar-reasoning", "unknown%2Fperplexity-sonar-reasoning"],
+  ["phi-3-medium-4k-instruct", "microsoft%2Fphi-3-medium-4k"],
+  ["phi-3-mini-128k-instruct", "microsoft%2Fphi-3-mini-128k"],
+  ["phi-3-mini-4k-instruct", "microsoft%2Fphi-3-mini-4k"],
+  ["phi-3-small-8k-instruct", "microsoft%2Fphi-3-small-8k"],
+  ["phi-4", "microsoft%2Fphi-4"],
+  ["prithivMLmods%2FPhi4-Super", "prithivMLmods%2FPhi-4-Super"],
+  ["qwen%2FQwen2.5-7B", "Qwen%2FQwen2.5-7B"],
+  ["qwen%2FQwen3-235B", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqvq-72b-preview", "Qwen%2FQVQ-72B-Preview"],
+  ["qwen%2Fqwen-2.5-14b-instruct", "Qwen%2FQwen2.5-14B"],
+  ["qwen%2Fqwen-2.5-32b-instruct", "Qwen%2FQwen2.5-32B"],
+  ["qwen%2Fqwen-2.5-72b-instruct", "Qwen%2FQwen2.5-72B"],
+  ["qwen%2Fqwen-2.5-7b-instruct", "Qwen%2FQwen2.5-7B"],
+  ["qwen%2Fqwen-2.5-coder-32b-instruct", "Qwen%2FQwen2.5-Coder-32B"],
+  ["qwen%2Fqwen-2.5-coder-7b-instruct", "Qwen%2FQwen2.5-Coder-7B"],
+  ["qwen%2Fqwen-2.5-vl-72b-instruct", "alibaba%2Fqwen2.5-vl-72b"],
+  ["qwen%2Fqwen1.5-110b-chat", "Qwen%2FQwen1.5-110B"],
+  ["qwen%2Fqwen1.5-14b", "Qwen%2FQwen1.5-14B"],
+  ["qwen%2Fqwen1.5-32b", "Qwen%2FQwen1.5-32B"],
+  ["qwen%2Fqwen1.5-72b", "Qwen%2FQwen1.5-72B"],
+  ["qwen%2Fqwen1.5-72b-chat", "Qwen%2FQwen1.5-72B"],
+  ["qwen%2Fqwen1.5-7b", "Qwen%2FQwen1.5-7B"],
+  ["qwen%2Fqwen2-72b-instruct", "Qwen%2FQwen2-72B"],
+  ["qwen%2Fqwen2-7b-instruct", "Qwen%2FQwen2-7B"],
+  ["qwen%2Fqwen2-vl-72b", "Qwen%2FQwen2-VL-72B"],
+  ["qwen%2Fqwen2.5-72b-instruct-turbo", "Qwen%2FQwen2.5-72B"],
+  ["qwen%2Fqwen2.5-7b-instruct-turbo", "Qwen%2FQwen2.5-7B"],
+  ["qwen%2Fqwen2.5-omni-7b", "Qwen%2FQwen2.5-Omni-7B"],
+  ["qwen%2Fqwen2.5-vl-72b", "alibaba%2Fqwen2.5-vl-72b"],
+  ["qwen%2Fqwen2.5-vl-7b", "alibaba%2Fqwen2.5-vl-7b"],
+  ["qwen%2Fqwen3-14b", "Qwen%2FQwen3-14B"],
+  ["qwen%2Fqwen3-14b-fc", "Qwen%2FQwen3-14B"],
+  ["qwen%2Fqwen3-14b-prompt", "Qwen%2FQwen3-14B"],
+  ["qwen%2Fqwen3-235b-a22b", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-2507", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-instruct-2507", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-instruct-2507-fc", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-instruct-2507-fp8", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-instruct-2507-prompt", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-235b-a22b-thinking-2507", "Qwen%2FQwen3-235B-A22B"],
+  ["qwen%2Fqwen3-30b-a3b", "Qwen%2FQwen3-30B-A3B"],
+  ["qwen%2Fqwen3-30b-a3b-instruct-2507", "Qwen%2FQwen3-30B-A3B"],
+  ["qwen%2Fqwen3-30b-a3b-instruct-2507-fc", "Qwen%2FQwen3-30B-A3B"],
+  ["qwen%2Fqwen3-30b-a3b-instruct-2507-prompt", "Qwen%2FQwen3-30B-A3B"],
+  ["qwen%2Fqwen3-32b", "Qwen%2FQwen3-32B"],
+  ["qwen%2Fqwen3-32b-fc", "Qwen%2FQwen3-32B"],
+  ["qwen%2Fqwen3-32b-nothink", "Qwen%2FQwen3-32B"],
+  ["qwen%2Fqwen3-32b-prompt", "Qwen%2FQwen3-32B"],
+  ["qwen%2Fqwen3-4b-instruct", "alibaba%2Fqwen3-4b"],
+  ["qwen%2Fqwen3-8b", "Qwen%2FQwen3-8B"],
+  ["qwen%2Fqwen3-8b-fc", "Qwen%2FQwen3-8B"],
+  ["qwen%2Fqwen3-8b-nothink", "Qwen%2FQwen3-8B"],
+  ["qwen%2Fqwen3-8b-prompt", "Qwen%2FQwen3-8B"],
+  ["qwen%2Fqwen3-coder-480b-a35b-instruct", "alibaba%2Fqwen3-coder-480b-a35b"],
+  ["qwen%2Fqwen3-max", "alibaba%2Fqwen3-max"],
+  ["qwen%2Fqwen3-next-80b-a3b-instruct", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["qwen%2Fqwen3-next-80b-a3b-thinking", "alibaba%2Fqwen3-next-80b-a3b"],
+  ["qwen%2Fqwen3-vl-235b-a22b-instruct", "alibaba%2Fqwen3-vl-235b-a22b"],
+  ["qwen%2Fqwen3-vl-235b-a22b-thinking", "alibaba%2Fqwen3-vl-235b-a22b"],
+  ["qwen%2Fqwen3-vl-30b-a3b-instruct", "alibaba%2Fqwen3-vl-30b-a3b"],
+  ["qwen%2Fqwen3-vl-30b-a3b-thinking", "alibaba%2Fqwen3-vl-30b-a3b"],
+  ["qwen%2Fqwen3-vl-32b-instruct", "Qwen%2FQwen3-VL-32B-Thinking"],
+  ["qwen%2Fqwen3-vl-32b-thinking", "Qwen%2FQwen3-VL-32B-Thinking"],
+  ["qwen%2Fqwen3-vl-4b-instruct", "alibaba%2Fqwen3-vl-4b"],
+  ["qwen%2Fqwen3-vl-4b-thinking", "alibaba%2Fqwen3-vl-4b"],
+  ["qwen%2Fqwen3-vl-8b-instruct", "Qwen%2FQwen3-VL-8B-Thinking"],
+  ["qwen%2Fqwen3-vl-8b-thinking", "Qwen%2FQwen3-VL-8B-Thinking"],
+  ["qwen%2Fqwen3.5-0.8b", "Qwen%2FQwen3.5-0.8B"],
+  ["qwen%2Fqwen3.5-122b", "alibaba%2Fqwen3-5-122b"],
+  ["qwen%2Fqwen3.5-122b-a10b", "alibaba%2Fqwen3-5-122b"],
+  ["qwen%2Fqwen3.5-122b-nothink", "alibaba%2Fqwen3-5-122b"],
+  ["qwen%2Fqwen3.5-27b", "Qwen%2FQwen3.5-27B"],
+  ["qwen%2Fqwen3.5-27b-nothink", "Qwen%2FQwen3.5-27B"],
+  ["qwen%2Fqwen3.5-2b", "Qwen%2FQwen3.5-2B"],
+  ["qwen%2Fqwen3.5-35b-a3b", "qwen%2Fqwen3.5-35b"],
+  ["qwen%2Fqwen3.5-397b-a17b", "alibaba%2Fqwen3-5-397b"],
+  ["qwen%2Fqwen3.5-4b", "Qwen%2FQwen3.5-4B"],
+  ["qwen%2Fqwen3.5-9b", "Qwen%2FQwen3.5-9B"],
+  ["qwen%2Fqwen3.6-27b", "Qwen%2FQwen3.6-27B"],
+  ["qwen%2Fqwen3.6-35b-a3b", "alibaba%2Fqwen3-6-35b"],
+  ["qwen%2Fqwen3.6-plus", "alibaba%2Fqwen3.6-plus"],
+  ["qwen%2Fqwq-32b", "Qwen%2FQwQ-32B"],
+  ["qwen1.5-0.5b-chat", "Qwen%2FQwen1.5-0.5B-Chat"],
+  ["qwen1.5-1.8b-chat", "Qwen%2FQwen1.5-1.8B-Chat"],
+  ["qwen1.5-110b-chat", "Qwen%2FQwen1.5-110B"],
+  ["qwen1.5-4b-chat", "Qwen%2FQwen1.5-4B-Chat"],
+  ["qwen1.5-72b-chat", "Qwen%2FQwen1.5-72B"],
+  ["qwen1.5-7b-chat", "Qwen%2FQwen1.5-7B"],
+  ["qwen2-72b-instruct", "Qwen%2FQwen2-72B"],
+  ["qwen2-math-72b-instruct", "Qwen%2FQwen2-Math-72B-Instruct"],
+  ["reflection-llama-3.1-70b", "mattshumer%2FReflection-Llama-3.1-70B"],
+  ["sarvam%2Fsarvam-105b", "sarvamai%2Fsarvam-105b"],
+  ["sarvam%2Fsarvam-30b", "sarvamai%2Fsarvam-30b"],
+  ["sarvam%2Fsarvam-m-reasoning", "sarvam%2Fsarvam-m"],
+  ["snowflake%2Farctic-instruct", "Snowflake%2Fsnowflake-arctic-instruct"],
+  ["snowflake%2Fsnowflake-arctic-instruct", "Snowflake%2Fsnowflake-arctic-instruct"],
+  ["sometimesanotion%2FLamarck-14B-v0.7-rc4", "sometimesanotion%2FLamarck-14B-v0.7"],
+  ["stabilityai%2Fstablelm-2-12b-chat", "stabilityai%2Fstablelm-2-12b"],
+  ["stabilityai%2Fstablelm-2-1_6b-chat", "stabilityai%2Fstablelm-2-1_6b"],
+  ["stanford%2Falpaca-7b", "stanford%2FAlpaca-7B"],
+  ["stepfun%2Fstep-3-5-flash", "stepfun-ai%2FStep-3.5-Flash"],
+  ["stepfun%2Fstep-3.5-flash", "stepfun-ai%2FStep-3.5-Flash"],
+  ["swiss-ai-initiative%2Fapertus-70b-instruct", "swiss-ai-initiative%2Fapertus-70b"],
+  ["swiss-ai-initiative%2Fapertus-8b-instruct", "swiss-ai-initiative%2Fapertus-8b"],
+  ["thudm%2Fchatglm-6b", "zai-org%2Fchatglm-6b"],
+  ["tiger-lab%2Fmammoth2-7b-plus", "TIGER-Lab%2FMAmmoTH2-7B-Plus"],
+  ["tiger-lab%2Fmammoth2-8b-plus", "TIGER-Lab%2FMAmmoTH2-8B-Plus"],
+  ["tiger-lab%2Fmammoth2-8x7b-plus", "TIGER-Lab%2FMAmmoTH2-8x7B-Plus"],
+  ["tii-uae%2Ffalcon3-10b-instruct-fc", "tiiuae%2Ffalcon3-10b"],
+  ["tii-uae%2Ffalcon3-1b-instruct-fc", "tii-uae%2Ffalcon3-1b"],
+  ["tii-uae%2Ffalcon3-3b-instruct-fc", "tii-uae%2Ffalcon3-3b"],
+  ["tii-uae%2Ffalcon3-7b-instruct-fc", "tiiuae%2Ffalcon3-7b"],
+  ["tiiuae%2FFalcon-40B", "tiiuae%2Ffalcon-40b"],
+  ["tiiuae%2FFalcon-7B", "tiiuae%2Ffalcon-7b"],
+  ["tiiuae%2FFalcon3-10B-Instruct", "tiiuae%2Ffalcon3-10b"],
+  ["tiiuae%2FFalcon3-1B-Instruct", "tii-uae%2Ffalcon3-1b"],
+  ["tiiuae%2FFalcon3-3B-Instruct", "tii-uae%2Ffalcon3-3b"],
+  ["tiiuae%2FFalcon3-7B-Instruct", "tiiuae%2Ffalcon3-7b"],
+  ["tiiuae%2Ffalcon-40b-instruct", "tiiuae%2Ffalcon-40b"],
+  ["tiiuae%2Ffalcon-7b-instruct", "tiiuae%2Ffalcon-7b"],
+  ["tinycompany%2FSigmaBoi-bgem3", "tinycompany%2FSigmaBoi-bge-m3"],
+  ["tinycompany%2FSigmaBoi-nomic-moe", "tinycompany%2FSigmaBoi-nomic-moe-bf16"],
+  ["together%2Fredpajama-incite-base-7b", "together%2FRedPajama-INCITE-Base-7B"],
+  ["together%2Fredpajama-incite-instruct-7b", "together%2FRedPajama-INCITE-Instruct-7B"],
+  ["togethercomputer%2FRedPajama-INCITE-7B-Base", "togethercomputer%2Fredpajama-incite-7b"],
+  ["togethercomputer%2FRedPajama-INCITE-7B-Chat", "togethercomputer%2Fredpajama-incite-7b"],
+  ["togethercomputer%2FRedPajama-INCITE-7B-Instruct", "togethercomputer%2Fredpajama-incite-7b"],
+  ["trillionlabs%2Ftri-21b-think-preview", "trillionlabs%2FTri-21B-Think-Preview"],
+  ["unknown%2Fdoubao-1.5-pro", "unknown%2FDoubao-1.5-pro"],
+  ["unknown%2Fiswe_agent", "unknown%2FiSWE_Agent"],
+  ["upstage%2Fsolar-open-100b-reasoning", "upstage%2Fsolar-open-100b"],
+  ["upstage%2Fsolar-pro-2", "upstage%2Fsolar-pro2"],
+  ["vicgalle%2FHumanish-RP-Llama-3.1-8B", "vicgalle%2FHumanish-Roleplay-Llama-3.1-8B"],
+  ["vicuna-7b-v1.5", "lmsys%2Fvicuna-7b-v1.5"],
+  ["wbot-4%3A347b_no_s", "wbot-4347b-no-s"],
+  ["weqweasdas%2FRM-Gemma-7B-4096", "weqweasdas%2FRM-Gemma-7B"],
+  ["writer%2Fpalmyra-x-004", "writer%2Fpalmyra-x"],
+  ["writer%2Fpalmyra-x-v2", "writer%2Fpalmyra-x"],
+  ["writer%2Fpalmyra-x-v3", "writer%2Fpalmyra-x"],
+  ["xai%2FGrok%204", "xai%2Fgrok-4"],
+  ["xai%2Fgrok-1", "xai-org%2Fgrok-1"],
+  ["xai%2Fgrok-2-1212", "xai%2Fgrok-2"],
+  ["xai%2Fgrok-2-mini", "xai%2Fgrok-2"],
+  ["xai%2Fgrok-2-vision-1212", "xai%2Fgrok-2"],
+  ["xai%2Fgrok-3-beta", "xai%2Fgrok-3"],
+  ["xai%2Fgrok-3-mini-beta", "xai%2Fgrok-3-mini"],
+  ["xai%2Fgrok-3-mini-beta-low-openrouter", "xai%2Fgrok-3-mini"],
+  ["xai%2Fgrok-3-mini-reasoning", "xai%2Fgrok-3-mini"],
+  ["xai%2Fgrok-3-openrouter", "xai%2Fgrok-3"],
+  ["xai%2Fgrok-4-0709", "xai%2Fgrok-4"],
+  ["xai%2Fgrok-4-0709-fc", "xai%2Fgrok-4"],
+  ["xai%2Fgrok-4-0709-prompt", "xai%2Fgrok-4"],
+  ["xai%2Fgrok-4-20-0309", "xai%2Fgrok-4"],
+  ["xai%2Fgrok-4-20-non-reasoning", "xai%2Fgrok-4.20-non-reasoning"],
+  ["xai%2Fgrok-4-heavy", "xai%2Fgrok-4"],
+  ["xiaomi%2Fmimo-v2-0206", "xiaomi%2Fmimo-v2"],
+  ["xiaomi%2Fmimo-v2-flash", "xiaomi%2Fmimo-v2"],
+  ["xiaomi%2Fmimo-v2-pro", "xiaomi%2Fmimo-v2"],
+  ["yandex%2FYaLM-100B", "yandex%2Fyalm-100b"],
+  ["yi-6b-chat", "01-ai%2FYi-6B"],
+  ["z-ai%2Fglm-4.5", "zai-org%2FGLM-4.5"],
+  ["zai%2Fglm-4-5-air", "zai-org%2FGLM-4.5-Air"],
+  ["zai%2Fglm-4-5v", "zai%2Fglm-4.5v"],
+  ["zai%2Fglm-4-5v-reasoning", "zai%2Fglm-4.5v"],
+  ["zai%2Fglm-4-6", "zai-org%2FGLM-4.6"],
+  ["zai%2Fglm-4-6-reasoning", "zai-org%2FGLM-4.6"],
+  ["zai%2Fglm-4-6v", "zai%2Fglm-4.6v"],
+  ["zai%2Fglm-4-6v-reasoning", "zai%2Fglm-4.6v"],
+  ["zai%2Fglm-4-7", "zai-org%2FGLM-4.7"],
+  ["zai%2Fglm-4-7-flash", "zai-org%2FGLM-4.7-Flash"],
+  ["zai%2Fglm-4-7-non-reasoning", "zai-org%2FGLM-4.7"],
+  ["zai%2Fglm-4.5", "zai-org%2FGLM-4.5"],
+  ["zai%2Fglm-4.6", "zai-org%2FGLM-4.6"],
+  ["zai%2Fglm-4.7", "zai-org%2FGLM-4.7"],
+  ["zai%2Fglm-5", "zai-org%2FGLM-5"],
+  ["zai%2Fglm-5-1", "zai-org%2FGLM-5.1"],
+  ["zai%2Fglm-5-1-non-reasoning", "zai-org%2FGLM-5.1"],
+  ["zai%2Fglm-5-non-reasoning", "zai-org%2FGLM-5"],
+  ["zai%2Fglm-5-thinking", "zai-org%2FGLM-5"],
+  ["zai%2Fglm-5.1-thinking", "zai-org%2FGLM-5.1"],
+  ["zai-org%2FGLM-4.6-FP8", "zai-org%2FGLM-4.6"],
+  ["zai-org%2FGLM-4.7-FP8", "zai-org%2FGLM-4.7"],
+  ["zai-org%2FGLM-5-FP8", "zai-org%2FGLM-5"],
+  ["zai-org%2FGLM-5.1-FP8", "zai-org%2FGLM-5.1"],
+  ["zai-org%2Fglm-4.5", "zai-org%2FGLM-4.5"],
+  ["zai-org%2Fglm-4.5-air", "zai-org%2FGLM-4.5-Air"],
+  ["zai-org%2Fglm-4.5-air-fp8", "zai-org%2FGLM-4.5-Air"],
+  ["zai-org%2Fglm-4.6", "zai-org%2FGLM-4.6"],
+  ["zai-org%2Fglm-4.7", "zai-org%2FGLM-4.7"],
+  ["zai-org%2Fglm-4.7-flash", "zai-org%2FGLM-4.7-Flash"],
+  ["zai-org%2Fglm-5", "zai-org%2FGLM-5"],
+  ["zai-org%2Fglm-5.1", "zai-org%2FGLM-5.1"],
+  ["zai-org%2Fglm-5v-turbo", "zai%2Fglm-5v-turbo"],
+  ["zelk12%2FTest01012025155054", "zelk12%2FTest01012025155054_gemma-2"],
+  ["zephyr-7b-alpha", "HuggingFaceH4%2Fzephyr-7b-alpha"],
+  ["zephyr-7b-beta", "HuggingFaceH4%2Fzephyr-7b-beta"],
+  ["zhipu%2FGLM%204.6", "zai-org%2FGLM-4.6"],
+  ["zhipu%2FGLM%204.7", "zai-org%2FGLM-4.7"],
+  ["zhipu%2Fglm-4.5", "zai-org%2FGLM-4.5"],
+  ["zhipu%2Fglm-5", "zai-org%2FGLM-5"],
+  ["zhipu-ai%2Fglm-4-9b", "zai-org%2Fglm-4-9b"],
+  ["zhipu-ai%2Fglm-4-9b-chat", "zai-org%2Fglm-4-9b"],
+  ["zhipu-ai%2Fglm-4.5", "zai-org%2FGLM-4.5"],
+  ["zhipu-ai%2Fglm-4.5-air", "zai-org%2FGLM-4.5-Air"],
+  ["zhipu-ai%2Fglm-4.6", "zai-org%2FGLM-4.6"],
+  ["zhipu-ai%2Fglm-4.7", "zai-org%2FGLM-4.7"],
+  ["zhipu-ai%2Fglm-5", "zai-org%2FGLM-5"],
+])
+
+export function resolveModelRedirect(routeId: string): string | undefined {
+  return MODEL_URL_REDIRECTS.get(routeId)
+}

lib/utils.ts

@@ -30,6 +30,22 @@ export function routeIdFromSegments(value: string | string[] | undefined): strin
   return joined.replace(/\//g, "%2F")
 }
 
+/**
+ * Build the backend `%2F`-encoded route id form from a plain canonical
+ * model id (`org/name`). Matches the producer's `route_id` /
+ * `model_route_id` encoding (RFC 3986 percent-encoding of the whole id),
+ * so the result can be fed straight into `routeIdToPath` for a URL path.
+ *
+ * This is the model-resolution-rework replacement for the old
+ * client-side family-route computation (since removed): the
+ * group/leaf id is now server-provided (`model_group_id`), and we only
+ * need to encode it for routing — never re-derive it.
+ */
+export function routeIdFromModelId(id: string | null | undefined): string {
+  if (!id) return ""
+  return encodeURIComponent(id.trim())
+}
+
 /**
  * Title-case a benchmark / family / eval label that arrives in slug or
  * snake-case form. Example inputs and outputs:

lib/view-data.ts

@@ -551,6 +551,8 @@ function reshapeCellToModelResult(row: Row): ModelResultForBenchmark {
   return {
     model_info: (modelInfo ?? modelInfoFromModelRow(row)) as ModelInfo,
     model_route_id: optionalString(row.model_route_id),
+    // model-resolution-rework: server-provided group id for routing fallback.
+    model_group_id: optionalString(row.model_group_id),
     score: scoreDetails.score,
     score_details: scoreDetails,
     evaluation_timestamp: asString(row.evaluation_timestamp, ""),
@@ -631,10 +633,16 @@ function modelSummaryFromRows(modelRow: Row, cellRows: Row[]): ModelEvaluationSu
     ...variant,
     variant_id: asString(variant.variant_id ?? variant.variant_key, `variant-${index}`),
     variant_key: asString(variant.variant_key, `variant-${index}`),
+    // Carry the GROUP's encoded route onto every variant. comparison-index /
+    // peer-ranks are keyed by the group route_id, and the model detail page
+    // renders the selected VARIANT as its summary — without this the variant
+    // has no model_route_id and the page falls back to a non-matching id, so
+    // peer-comparison charts find no scores.
+    model_route_id: asString(modelRow.model_route_id ?? modelRow.route_id, modelRow.route_id),
     variant_label: asString(variant.variant_label ?? variant.variant_display_name, "Default"),
     variant_display_name: asString(variant.variant_display_name ?? variant.variant_label ?? modelRow.model_name, modelRow.model_name),
     raw_model_ids: asArray<string>(variant.raw_model_ids),
-    family_id: asString(variant.family_id ?? modelRow.model_family_id, modelRow.model_family_id),
+    family_id: asString(variant.family_id ?? modelRow.model_group_id, modelRow.model_group_id),
     family_name: asString(variant.family_name ?? modelRow.model_family_name, modelRow.model_family_name),
     total_evaluations: asNumber(variant.total_evaluations ?? totalEvaluations),
     last_updated: asString(variant.last_updated ?? lastUpdated, lastUpdated),
@@ -649,10 +657,17 @@ function modelSummaryFromRows(modelRow: Row, cellRows: Row[]): ModelEvaluationSu
 
   return {
     ...core,
-    model_family_id: asString(modelRow.model_family_id ?? modelRow.model_key ?? modelRow.model_id, modelRow.model_key ?? modelRow.model_id),
+    model_group_id: asString(modelRow.model_group_id ?? modelRow.model_key ?? modelRow.model_id, modelRow.model_key ?? modelRow.model_id),
     model_route_id: asString(modelRow.model_route_id ?? modelRow.route_id, modelRow.route_id),
     model_family_name: asString(modelRow.model_family_name ?? modelRow.model_name, modelRow.model_name),
     raw_model_ids: rawModelIds.length > 0 ? rawModelIds : [asString(modelRow.model_key ?? modelRow.model_id, "")].filter(Boolean),
+    // model-resolution-rework (additive, nullable). The summary builder
+    // reads `SELECT *` from models_view, so these columns flow through
+    // once the producer view layer emits them (post-M9). Until then
+    // optionalString yields undefined and the UI conditionally omits them.
+    lineage_origin_model_id: optionalString(modelRow.lineage_origin_model_id),
+    resolution_source: optionalString(modelRow.resolution_source),
+    resolution_granularity: optionalString(modelRow.resolution_granularity),
     variants,
   }
 }
@@ -751,7 +766,7 @@ export async function getDashboardData() {
 
 export async function getModelSummaryById(routeId: string): Promise<ModelEvaluationSummary | null> {
   // Lookups use the addressable identifier (`model_key`/`route_id`/
-  // `model_route_id`/`model_family_id`) so unresolved models — whose
+  // `model_route_id`/`model_group_id`) so unresolved models — whose
   // `model_id` is NULL — are still findable. `model_id` is kept in the
   // OR chain as a back-compat fallback for old links.
   //
@@ -760,20 +775,38 @@ export async function getModelSummaryById(routeId: string): Promise<ModelEvaluat
   //     Next.js already decodes path params before they reach here, so
   //     `routeId` lands as `google/gemini-3-pro`.
   //   - Plain canonical id with `/` (same shape after Next.js decode).
-  //   - Legacy `__`-separated form (e.g. `google__gemini-3-pro`) — old
-  //     `getModelFamilyRouteId` emitted this; bookmarks may still use
-  //     it. Convert `__` → `/` for lookup.
+  //   - Legacy `__`-separated form (e.g. `google__gemini-3-pro`) — the
+  //     old client-side family route computation emitted this; bookmarks
+  //     may still use it. Convert `__` → `/` for lookup.
   const dunder = routeId.includes("__") ? routeId.replace(/__/g, "/") : routeId
   const rows = await readRows<Row>(
     `SELECT *
      FROM models_view
-     WHERE model_key = ? OR route_id = ? OR model_route_id = ? OR model_family_id = ? OR model_id = ?
+     WHERE model_key = ? OR route_id = ? OR model_route_id = ? OR model_group_id = ? OR model_id = ?
         OR model_key = ? OR model_id = ?
      LIMIT 1`,
     [routeId, routeId, routeId, routeId, routeId, dunder, dunder],
     { contextLabel: `model_lookup=${routeId}` }
   )
-  const modelRow = rows[0]
+  let modelRow = rows[0]
+  if (!modelRow) {
+    // The id may be a FOLDED raw id — a dated snapshot / older-cased / variant
+    // spelling the producer collapsed into a group (e.g.
+    // `mistralai/mistral-medium-2505` folds into `mistralai/mistral-medium`).
+    // Such ids exist only inside the owning group row's `raw_model_ids` list,
+    // never as their own row, so resolve them to that group. Case-insensitive,
+    // since raw_model_ids preserve HF casing. This makes every inbound model
+    // link/bookmark resolve to a real page instead of 404-ing.
+    const byRaw = await readRows<Row>(
+      `SELECT *
+       FROM models_view
+       WHERE list_contains(list_transform(raw_model_ids, x -> lower(x)), lower(?))
+          OR list_contains(list_transform(raw_model_ids, x -> lower(x)), lower(?))
+       LIMIT 1`,
+      [routeId, dunder]
+    )
+    modelRow = byRaw[0]
+  }
   if (!modelRow) return null
 
   const cellRows = await getModelEvaluationRows(asString(modelRow.model_key ?? modelRow.model_id, routeId))
@@ -888,6 +921,7 @@ export async function getEvalSummaryById(evalId: string): Promise<BenchmarkEvalS
         return {
           model_info: base.model_info,
           model_route_id: row.model_route_id,
+          model_group_id: base.model_group_id,
           evaluation_timestamp: base.evaluation_timestamp,
           source_metadata: base.source_metadata,
           source_data: base.source_data,
@@ -939,6 +973,7 @@ export async function getEvalSummaryById(evalId: string): Promise<BenchmarkEvalS
       .map((mr) => ({
         model_info: mr.model_info,
         model_route_id: mr.model_route_id,
+        model_group_id: mr.model_group_id,
         evaluation_timestamp: mr.evaluation_timestamp,
         source_metadata: mr.source_metadata,
         source_data: mr.source_data,
@@ -962,9 +997,25 @@ export async function getDeveloperList(): Promise<DeveloperListEntry[]> {
   return [...(headline.developers ?? [])].sort((a, b) => a.developer.localeCompare(b.developer))
 }
 
+function decodeLoose(value: string): string {
+  try {
+    return decodeURIComponent(value)
+  } catch {
+    return value
+  }
+}
+
 export async function getDeveloperSummaryById(routeId: string) {
   const developers = await getDeveloperList()
-  const developer = developers.find((entry) => entry.route_id === routeId)
+  // `route_id` is stored pre-percent-encoded for names with spaces/parens
+  // (e.g. "Mistral AI" -> "Mistral%20AI"), but the incoming routeId arrives
+  // already decoded (routeIdFromSegments only round-trips %2F). Compare on the
+  // decoded form so those developers' detail pages resolve. Exact match is
+  // tried first as a fast path.
+  const target = decodeLoose(routeId)
+  const developer =
+    developers.find((entry) => entry.route_id === routeId) ??
+    developers.find((entry) => decodeLoose(entry.route_id) === target)
   if (!developer) return null
 
   const modelRows = await readRows<Row>(

middleware.ts

@@ -0,0 +1,70 @@
+// model-resolution-rework — 301 redirects for changed model URLs.
+//
+// ~297 model routes change at the registry re-baseline (204 canonical_id
+// flips root->leaf + 93 casing re-keys). Bookmarked / inbound links to the
+// OLD route must 301 to the NEW one.
+//
+// WHY middleware and not `app/models/[...id]/route.ts`: a Route Handler
+// (`route.ts`) cannot coexist with the existing `page.tsx` in the same App
+// Router segment — Next.js treats that as a conflict. Middleware runs before
+// routing, sees every `/models/*` request, and can issue a 301 with full
+// control over the URL and query string. (Deviation from the playbook's
+// suggested file path; same behaviour, correct Next.js mechanism.)
+//
+// CONTRACT (tested in tests/transformations/model-url-redirects.test.ts):
+//   - 301 (permanent) so caches/search engines update.
+//   - QUERY PARAMS PRESERVED across the redirect (e.g. ?version=...), so a
+//     deep link to a specific variant survives.
+//   - the model id round-trips through percent-encoding: the incoming path
+//     segment is decoded by Next.js, we re-encode to the producer `route_id`
+//     form (whole-id encoding, `/` -> %2F) to match the redirect-map keys.
+
+import { NextResponse } from "next/server"
+import type { NextRequest } from "next/server"
+
+import { resolveModelRedirect } from "@/lib/model-url-redirects"
+
+export const config = {
+  // Only run on model detail routes. Static assets / api are untouched.
+  matcher: ["/models/:path*"],
+}
+
+const MODELS_PREFIX = "/models/"
+
+export function middleware(req: NextRequest): NextResponse | undefined {
+  const { pathname } = req.nextUrl
+  if (!pathname.startsWith(MODELS_PREFIX)) return undefined
+
+  // Everything after `/models/`. Next.js has already percent-DECODED the
+  // path, so `/models/anthropic%2Fclaude-3-haiku` arrives here as the
+  // decoded `anthropic/claude-3-haiku`. Re-encode the whole id to the
+  // producer `route_id` form to look up the map.
+  const rawTail = pathname.slice(MODELS_PREFIX.length)
+  if (!rawTail) return undefined
+
+  const decoded = safeDecode(rawTail)
+  const routeId = encodeURIComponent(decoded)
+
+  const newRouteId = resolveModelRedirect(routeId)
+  if (!newRouteId) return undefined
+
+  // Build the target. encodeURIComponent gives the whole-id `%2F` form;
+  // keep it as a single path segment under the [...id] catch-all so the
+  // page resolves it exactly like a freshly-minted link.
+  const target = req.nextUrl.clone()
+  target.pathname = `${MODELS_PREFIX}${newRouteId}`
+  // PRESERVE query params (e.g. ?version=20240620). `clone()` already copies
+  // search; set it explicitly so intent is obvious and robust to refactors.
+  target.search = req.nextUrl.search
+
+  return NextResponse.redirect(target, 301)
+}
+
+function safeDecode(value: string): string {
+  try {
+    return decodeURIComponent(value)
+  } catch {
+    // Malformed percent-encoding — treat the raw value as already-decoded.
+    return value
+  }
+}

package.json

@@ -12,6 +12,8 @@
     "start": "next start",
     "test": "vitest",
     "test:drift": "RUN_DRIFT=1 vitest --run tests/upstream-drift.test.ts",
+    "test:integrity": "vitest --run tests/redirect-integrity.test.ts",
+    "test:e2e": "playwright test",
     "compare-data-backends": "node scripts/compare-data-backends.mjs",
     "refresh-fixtures": "node scripts/refresh-fixtures.mjs",
     "audit-adapters": "tsx scripts/audit-adapters.mjs"
@@ -72,10 +74,12 @@
     "zod": "3.25.67"
   },
   "devDependencies": {
+    "@playwright/test": "^1.60.0",
     "@tailwindcss/postcss": "^4.1.9",
     "@types/node": "^22",
     "@types/react": "^19",
     "@types/react-dom": "^19",
+    "playwright": "^1.60.0",
     "postcss": "^8.5",
     "puppeteer-core": "^24.42.0",
     "tailwindcss": "^4.1.9",

playwright.config.ts

@@ -0,0 +1,30 @@
+import { defineConfig } from "@playwright/test"
+
+// Frontend preflight e2e (opt-in): `SNAPSHOT_URL=<warehouse> pnpm test:e2e`.
+// NOT part of `pnpm test` and nothing gates deploy — it's a correctness check you
+// run before shipping a meaningful frontend change. See tests/PREFLIGHT.md.
+//
+// The webServer block boots `pnpm dev` against the snapshot itself (and reuses an
+// already-running one), so there's no manual server dance. When SNAPSHOT_URL is
+// unset the spec self-skips.
+const SNAPSHOT_URL = process.env.SNAPSHOT_URL
+const PORT = Number(process.env.PORT || 3211)
+
+export default defineConfig({
+  testDir: "tests/e2e",
+  // Each test sweeps many pages; give them room (they parallelize internally).
+  timeout: 300_000,
+  fullyParallel: false,
+  workers: 1,
+  reporter: "list",
+  use: { baseURL: `http://localhost:${PORT}` },
+  webServer: SNAPSHOT_URL
+    ? {
+        command: `pnpm dev -p ${PORT}`,
+        url: `http://localhost:${PORT}`,
+        timeout: 120_000,
+        reuseExistingServer: true,
+        env: { DATA_BACKEND: "v2", SNAPSHOT_URL },
+      }
+    : undefined,
+})

pnpm-lock.yaml

@@ -115,7 +115,7 @@ importers:
         version: 8.5.1(react@19.1.1)
       geist:
         specifier: ^1.3.1
-        version: 1.4.2(next@15.2.8(react-dom@19.1.1(react@19.1.1))(react@19.1.1))
+        version: 1.4.2(next@15.2.8(@playwright/test@1.60.0)(react-dom@19.1.1(react@19.1.1))(react@19.1.1))
       input-otp:
         specifier: 1.4.1
         version: 1.4.1(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
@@ -124,7 +124,7 @@ importers:
         version: 0.454.0(react@19.1.1)
       next:
         specifier: 15.2.8
-        version: 15.2.8(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
+        version: 15.2.8(@playwright/test@1.60.0)(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
       next-themes:
         specifier: latest
         version: 0.4.6(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
@@ -168,6 +168,9 @@ importers:
         specifier: 3.25.67
         version: 3.25.67
     devDependencies:
+      '@playwright/test':
+        specifier: ^1.60.0
+        version: 1.60.0
       '@tailwindcss/postcss':
         specifier: ^4.1.9
         version: 4.1.12
@@ -180,6 +183,9 @@ importers:
       '@types/react-dom':
         specifier: ^19
         version: 19.1.7(@types/react@19.1.10)
+      playwright:
+        specifier: ^1.60.0
+        version: 1.60.0
       postcss:
         specifier: ^8.5
         version: 8.5.6
@@ -754,6 +760,11 @@ packages:
     cpu: [x64]
     os: [win32]
 
+  '@playwright/test@1.60.0':
+    resolution: {integrity: sha512-O71yZIbAh/PxDMNGns37GHBIfrVkEVyn+AXyIa5dOTfb4/xNvRWV+Vv/NMbNCtODB/pO7vLlF2OTmMVLhmr7Ag==}
+    engines: {node: '>=18'}
+    hasBin: true
+
   '@puppeteer/browsers@2.13.0':
     resolution: {integrity: sha512-46BZJYJjc/WwmKjsvDFykHtXrtomsCIrwYQPOP7VfMJoZY2bsDF9oROBABR3paDjDcmkUye1Pb1BqdcdiipaWA==}
     engines: {node: '>=18'}
@@ -2058,6 +2069,11 @@ packages:
   fraction.js@4.3.7:
     resolution: {integrity: sha512-ZsDfxO51wGAXREY55a7la9LScWpwv9RxIrYABrlvOFBlH/ShPnrtsXeuUIfXKKOVicNxQ+o8JTbJvjS4M89yew==}
 
+  fsevents@2.3.2:
+    resolution: {integrity: sha512-xiqMQR4xAeHTuB9uWm+fFRcIOgKBMiOBP+eXiyT7jsgVCq1bkVygt00oASowB7EdtpOHaaPgKt812P9ab+DDKA==}
+    engines: {node: ^8.16.0 || ^10.6.0 || >=11.0.0}
+    os: [darwin]
+
   fsevents@2.3.3:
     resolution: {integrity: sha512-5xoDfX+fL7faATnagmWPpbFtwh/R77WmMMqqHGS65C3vvB0YHrgF+B1YmZ3441tMj5n63k0212XNoJwzlhffQw==}
     engines: {node: ^8.16.0 || ^10.6.0 || >=11.0.0}
@@ -2351,6 +2367,16 @@ packages:
   pkg-types@1.3.1:
     resolution: {integrity: sha512-/Jm5M4RvtBFVkKWRu2BLUTNP8/M2a+UwuAX+ae4770q1qVGtfjG+WTCupoZixokjmHiry8uI+dlY8KXYV5HVVQ==}
 
+  playwright-core@1.60.0:
+    resolution: {integrity: sha512-9bW6zvX/m0lEbgTKJ6YppOKx8H3VOPBMOCFh2irXFOT4BbHgrx5hPjwJYLT40Lu+4qtD36qKc/Hn56StUW57IA==}
+    engines: {node: '>=18'}
+    hasBin: true
+
+  playwright@1.60.0:
+    resolution: {integrity: sha512-hheHdokM8cdqCb0lcE3s+zT4t4W+vvjpGxsZlDnikarzx8tSzMebh3UiFtgqwFwnTnjYQcsyMF8ei2mCO/tpeA==}
+    engines: {node: '>=18'}
+    hasBin: true
+
   postcss-value-parser@4.2.0:
     resolution: {integrity: sha512-1NNCs6uurfkVbeXG4S8JFT9t19m45ICnif8zWLd5oPSZ50QnwMfK+H3jv408d4jw/7Bttv5axS5IiHoLaVNHeQ==}
 
@@ -3177,6 +3203,10 @@ snapshots:
   '@next/swc-win32-x64-msvc@15.2.5':
     optional: true
 
+  '@playwright/test@1.60.0':
+    dependencies:
+      playwright: 1.60.0
+
   '@puppeteer/browsers@2.13.0':
     dependencies:
       debug: 4.4.3
@@ -4495,12 +4525,15 @@ snapshots:
 
   fraction.js@4.3.7: {}
 
+  fsevents@2.3.2:
+    optional: true
+
   fsevents@2.3.3:
     optional: true
 
-  geist@1.4.2(next@15.2.8(react-dom@19.1.1(react@19.1.1))(react@19.1.1)):
+  geist@1.4.2(next@15.2.8(@playwright/test@1.60.0)(react-dom@19.1.1(react@19.1.1))(react@19.1.1)):
     dependencies:
-      next: 15.2.8(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
+      next: 15.2.8(@playwright/test@1.60.0)(react-dom@19.1.1(react@19.1.1))(react@19.1.1)
 
   get-caller-file@2.0.5: {}
 
@@ -4664,7 +4697,7 @@ snapshots:
       react: 19.1.1
       react-dom: 19.1.1(react@19.1.1)
 
-  next@15.2.8(react-dom@19.1.1(react@19.1.1))(react@19.1.1):
+  next@15.2.8(@playwright/test@1.60.0)(react-dom@19.1.1(react@19.1.1))(react@19.1.1):
     dependencies:
       '@next/env': 15.2.8
       '@swc/counter': 0.1.3
@@ -4684,6 +4717,7 @@ snapshots:
       '@next/swc-linux-x64-musl': 15.2.5
       '@next/swc-win32-arm64-msvc': 15.2.5
       '@next/swc-win32-x64-msvc': 15.2.5
+      '@playwright/test': 1.60.0
       sharp: 0.33.5
     transitivePeerDependencies:
       - '@babel/core'
@@ -4747,6 +4781,14 @@ snapshots:
       mlly: 1.7.4
       pathe: 2.0.3
 
+  playwright-core@1.60.0: {}
+
+  playwright@1.60.0:
+    dependencies:
+      playwright-core: 1.60.0
+    optionalDependencies:
+      fsevents: 2.3.2
+
   postcss-value-parser@4.2.0: {}
 
   postcss@8.4.31:

scripts/generate-model-redirects.ts

@@ -0,0 +1,88 @@
+#!/usr/bin/env tsx
+// Generate lib/model-url-redirects.ts from the warehouse models_view.
+//
+// The producer folds variants / dated snapshots / older-cased spellings into a
+// group whose model_route_id is the addressable page; the folded spellings live
+// only in raw_model_ids. This emits a folded-spelling -> group-route 301 map so
+// a link to any folded spelling lands on the real group page. Targets are always
+// addressable by construction, so the map can never 301 to a dead URL.
+//
+// Usage:
+//   DATA_BACKEND=v2 SNAPSHOT_URL=<warehouse-url-or-file> tsx scripts/generate-model-redirects.ts
+//   tsx scripts/generate-model-redirects.ts --snapshot <warehouse-url> --out lib/model-url-redirects.ts
+
+import { writeFileSync } from "node:fs"
+import { dirname, resolve } from "node:path"
+import { fileURLToPath } from "node:url"
+
+import { DuckDBConnection } from "@duckdb/node-api"
+
+import {
+  buildRedirectsFromModelsView,
+  serializeRedirectModule,
+  type ModelsViewRow,
+} from "../lib/model-url-redirects-build"
+
+const REPO_ROOT = resolve(dirname(fileURLToPath(import.meta.url)), "..")
+const DEFAULT_OUT = resolve(REPO_ROOT, "lib/model-url-redirects.ts")
+
+function parseArg(flag: string): string | undefined {
+  const i = process.argv.indexOf(flag)
+  return i >= 0 ? process.argv[i + 1] : undefined
+}
+
+// @duckdb/node-api returns LIST columns as { items: [...] }.
+function listItems(v: unknown): string[] {
+  if (Array.isArray(v)) return v.map(String)
+  if (v && typeof v === "object" && Array.isArray((v as { items?: unknown[] }).items)) {
+    return (v as { items: unknown[] }).items.map(String)
+  }
+  return []
+}
+
+function str(v: unknown): string | null {
+  return v == null ? null : String(v)
+}
+
+async function main() {
+  const snapshot = (parseArg("--snapshot") ?? process.env.SNAPSHOT_URL ?? "").replace(/\/+$/, "")
+  if (!snapshot) {
+    throw new Error("need --snapshot <warehouse-url> or SNAPSHOT_URL set")
+  }
+  const outPath = resolve(parseArg("--out") ?? DEFAULT_OUT)
+
+  const con = await DuckDBConnection.create()
+  await con.run("INSTALL httpfs; LOAD httpfs;")
+  const reader = await con.runAndReadAll(
+    `SELECT model_route_id, model_id, route_id, model_key, model_group_id, raw_model_ids
+     FROM read_parquet('${snapshot}/models_view.parquet')`,
+  )
+  const raw = reader.getRowObjects() as Array<Record<string, unknown>>
+  const rows: ModelsViewRow[] = raw.map((r) => ({
+    model_route_id: String(r.model_route_id ?? ""),
+    model_id: str(r.model_id),
+    route_id: str(r.route_id),
+    model_key: str(r.model_key),
+    model_group_id: str(r.model_group_id),
+    raw_model_ids: listItems(r.raw_model_ids),
+  }))
+
+  const { redirects, ambiguous } = buildRedirectsFromModelsView(rows)
+  writeFileSync(outPath, serializeRedirectModule(redirects, { source: snapshot }))
+
+  // eslint-disable-next-line no-console
+  console.log(
+    `wrote ${outPath}: ${redirects.size} redirects from ${rows.length} models_view rows` +
+      (ambiguous.size ? `; ${ambiguous.size} ambiguous spellings excluded` : ""),
+  )
+  if (ambiguous.size > 0) {
+    // eslint-disable-next-line no-console
+    console.log(`  ambiguous (fan out to >1 group): ${[...ambiguous.keys()].slice(0, 10).join(", ")}`)
+  }
+}
+
+main().catch((e) => {
+  // eslint-disable-next-line no-console
+  console.error(e)
+  process.exit(1)
+})

scripts/verify-identity.mjs

@@ -1,36 +0,0 @@
-import "./server-only-shim.mjs"
-import fs from "fs"
-
-const { getCanonicalModelIdentity, getModelFamilyRouteId } = await import("../lib/model-family.ts")
-
-const cards = JSON.parse(fs.readFileSync(".cache/hf-data/model-cards.json", "utf8"))
-
-let total = 0
-let familyIdMismatch = 0
-let familyNameMismatch = 0
-const familyIdExamples = []
-const familyNameExamples = []
-
-for (const c of cards) {
-  total++
-  const computed = getCanonicalModelIdentity({
-    id: c.model_family_id,
-    name: c.model_family_name,
-  })
-  if (computed.familyId !== c.model_family_id) {
-    familyIdMismatch++
-    if (familyIdExamples.length < 5) familyIdExamples.push({pipeline: c.model_family_id, computed: computed.familyId})
-  }
-  if (computed.familyName !== c.model_family_name) {
-    familyNameMismatch++
-    if (familyNameExamples.length < 10) familyNameExamples.push({pipeline: c.model_family_name, computed: computed.familyName})
-  }
-  const computedRoute = getModelFamilyRouteId(computed.familyId)
-  if (computedRoute !== c.model_route_id) {
-    console.error("ROUTE MISMATCH", c.model_family_id, "->", computedRoute, "vs", c.model_route_id)
-  }
-}
-
-console.log({total, familyIdMismatch, familyNameMismatch})
-if (familyIdExamples.length) console.log("familyId examples:", familyIdExamples)
-if (familyNameExamples.length) console.log("familyName examples:", familyNameExamples.slice(0, 10))

scripts/verify-setup-alias.mjs

@@ -1,99 +0,0 @@
-import "./server-only-shim.mjs"
-import fs from "fs"
-
-const { getCanonicalModelIdentity, getModelFamilyRouteId } = await import("../lib/model-family.ts")
-
-// Replicate the runtime normalizer's logic exactly (lib/hf-data.ts:750-812)
-// so we can trace what it would do against any real card.
-function normalizeSetupAliasQualifier(value) {
-  return value?.trim().toLowerCase().replace(/[_\s]+/g, "-") ?? ""
-}
-function isSetupAliasQualifier(value) {
-  const normalized = normalizeSetupAliasQualifier(value)
-  return (
-    normalized === "prompt" ||
-    normalized === "fc" ||
-    normalized === "function-calling" ||
-    normalized.startsWith("thinking")
-  )
-}
-function getLatestTimestamp(a, b) {
-  if (!a) return b
-  if (!b) return a
-  const aTime = new Date(a).getTime()
-  const bTime = new Date(b).getTime()
-  if (!Number.isFinite(aTime)) return b
-  if (!Number.isFinite(bTime)) return a
-  return bTime > aTime ? b : a
-}
-function normalizeSingleModelCardEntry(entry) {
-  const familyIdentity = getCanonicalModelIdentity({ id: entry.model_family_id, name: entry.model_family_name })
-  const variantsByKey = new Map()
-  for (const variant of entry.variants ?? []) {
-    let normalizedVariantKey = variant.variant_key
-    let normalizedVariantLabel = variant.variant_label
-    if (variant.variant_key === "base") {
-      normalizedVariantKey = "default"
-      normalizedVariantLabel = "Default"
-    } else if (variant.variant_key !== "default") {
-      const syntheticIdentity = getCanonicalModelIdentity({
-        id: `${familyIdentity.familyId}-${variant.variant_key}`,
-        name: `${familyIdentity.familyId}-${variant.variant_key}`,
-      })
-      if (syntheticIdentity.versionDate && isSetupAliasQualifier(syntheticIdentity.versionQualifier)) {
-        normalizedVariantKey = syntheticIdentity.versionDate
-        normalizedVariantLabel = syntheticIdentity.versionDate
-      } else {
-        normalizedVariantKey = syntheticIdentity.variantKey
-        normalizedVariantLabel = syntheticIdentity.variantLabel
-      }
-    }
-    const existing = variantsByKey.get(normalizedVariantKey)
-    if (existing) {
-      existing.evaluation_count += variant.evaluation_count
-      existing.last_updated = getLatestTimestamp(existing.last_updated, variant.last_updated)
-      existing.raw_model_ids = Array.from(new Set([...(existing.raw_model_ids ?? []), ...(variant.raw_model_ids ?? [])])).sort()
-      continue
-    }
-    variantsByKey.set(normalizedVariantKey, {
-      ...variant,
-      variant_key: normalizedVariantKey,
-      variant_label: normalizedVariantLabel,
-      raw_model_ids: [...(variant.raw_model_ids ?? [])].sort(),
-    })
-  }
-  const normalizedVariants = Array.from(variantsByKey.values())
-  return { ...entry, variants: normalizedVariants }
-}
-
-const cards = JSON.parse(fs.readFileSync(".cache/hf-data/model-cards.json", "utf8"))
-
-// === Sample a known-affected card to show pre/post ===
-console.log("=== Sample: openai/gpt-5.2 (multi-variant flagship, known to exercise the rule) ===")
-const gpt52 = cards.find(c => c.model_family_id === "openai/gpt-5.2")
-if (gpt52) {
-  console.log("Input variants (cache state):")
-  for (const v of gpt52.variants) console.log(`  '${v.variant_key}'`)
-  console.log("After runtime normalize (what /api/model-cards returns):")
-  const normalized = normalizeSingleModelCardEntry(gpt52)
-  for (const v of normalized.variants) console.log(`  '${v.variant_key}' / raw_ids count=${v.raw_model_ids?.length ?? 0}`)
-}
-
-// === Aggregate audit: across all cards, does normalizer change anything? ===
-console.log("\n=== Aggregate: cards where normalizer would CHANGE the variants list ===")
-let changedCount = 0
-const changedExamples = []
-for (const c of cards) {
-  const beforeKeys = (c.variants ?? []).map(v => v.variant_key).sort().join("|")
-  const afterKeys = normalizeSingleModelCardEntry(c).variants.map(v => v.variant_key).sort().join("|")
-  if (beforeKeys !== afterKeys) {
-    changedCount++
-    if (changedExamples.length < 5) changedExamples.push({ family: c.model_family_id, before: beforeKeys, after: afterKeys })
-  }
-}
-console.log(`  ${changedCount} of ${cards.length} cards would have variants changed by runtime normalizer`)
-for (const ex of changedExamples) {
-  console.log(`  ${ex.family}:`)
-  console.log(`    before: ${ex.before}`)
-  console.log(`    after:  ${ex.after}`)
-}

test_redirects_manual.mjs

@@ -0,0 +1,23 @@
+import { resolveModelRedirect } from "./lib/model-url-redirects.ts"
+import { routeIdFromModelId } from "./lib/utils.ts"
+
+// Test 1: Percent-encoding round-trip with special characters
+const testIds = [
+  "anthropic/claude-3.5-sonnet-20240620",  // has '.'
+  "meta/Llama-3.1-8B",                      // has '.' and mixed case
+  "01-ai/Yi-1.5-34B",                       // has '.' and '-' and digits
+  "openai/gpt-4o",                          // lowercase with special combo
+]
+
+console.log("=== TEST 1: Percent-encoding round-trip ===")
+testIds.forEach(id => {
+  const encoded = routeIdFromModelId(id)
+  const decoded = decodeURIComponent(encoded)
+  const roundtrip = encoded.includes("%2F") ? "✓" : "✗ (slash not encoded!)"
+  const match = decoded === id ? "✓" : "✗ (mismatch!)"
+  console.log(`  ${id}`)
+  console.log(`    -> encoded: ${encoded} ${roundtrip}`)
+  console.log(`    -> decoded: ${decoded} ${match}`)
+})
+
+console.log("\n=== All manual tests completed ===")

tests/PREFLIGHT.md

@@ -0,0 +1,76 @@
+# Frontend preflight — correctness checks before a deploy
+
+Run these before pushing a meaningful frontend change to the HF Space. They are
+**opt-in and non-blocking** — nothing here gates `git push` or the default
+`pnpm test`; they're a correctness gate you choose to run.
+
+Guiding principle, learned the hard way: **"the page returns 200 / an SVG exists /
+no console error" ≠ correct.** A model page can render cleanly and still be wrong —
+a folded id 404s its data, a redirect points the wrong way, or a chart silently
+shows only the current model. These checks assert content, not just liveness.
+
+## How to run
+
+```bash
+export SNAPSHOT_URL="https://huggingface.co/datasets/evaleval/card_backend/resolve/main/warehouse/<id>"
+#   (or file:///abs/path/to/warehouse/<id>)
+
+# 1. Server-free data-contract checks (fast). Self-skips if SNAPSHOT_URL is unset.
+pnpm test:integrity        # tests/redirect-integrity.test.ts (vitest)
+
+# 2. Live-server page + chart-content checks (Playwright boots its own dev server).
+pnpm test:e2e              # tests/e2e/frontend-preflight.spec.ts
+```
+
+Both need `SNAPSHOT_URL`. `test:e2e` starts `pnpm dev` against it automatically
+(and reuses an already-running server on the port). Set `PORT` to override 3211.
+
+## What each layer asserts
+
+**`test:integrity` (vitest, server-free)** — against the warehouse snapshot:
+- redirect map has 0 self-redirects, 0 chains/loops, every target is an addressable route;
+- direction is folded→group (no addressable id is a redirect KEY — catches an inverted map);
+- every redirect key is a known folded `raw_model_id`;
+- no `raw_model_id` belongs to >1 group (the resolver's `LIMIT 1` fallback is unambiguous);
+- no `models_view` row has NULL `raw_model_ids`.
+
+**`test:e2e` (Playwright, live server)** — against the running app + `/api/comparison-index` ground truth:
+- model / eval / developer pages render with HTTP <400, no client error text, no console errors —
+  **including the regression sets**: folded model ids (from `raw_model_ids`) and percent-encoded
+  developer names (e.g. `Mistral AI`);
+- **100% of folded model ids resolve** (cheap status sweep — not a sample);
+- comparison charts render real **peer bars** — counted in the DOM via `data-model-bar` /
+  `data-bar-current`, NOT via `#n/m` badges (those come from the peer-*ranks* sidecar) or the
+  "No peer scores" string (which never fires on the silent single-bar failure). Catches the
+  "chart shows only the current model" bug. Also: 0 "Unknown Model" labels; the current model is
+  not rendered as its own peer.
+
+## The bug taxonomy these were built to catch
+
+1. Folded model ids 404 — lookup didn't match `raw_model_ids`.
+2. Inverted redirect map 301'd working group URLs to dead leaves.
+3. Percent-encoded developer names (`Mistral%20AI`) 404'd (decoded-vs-encoded mismatch).
+4. Comparison charts showed only the current model (peer lookup keyed on a non-matching id).
+5. Peer bars showed "Unknown Model" (label read a field that isn't on the score rows).
+
+The chart checks depend on the inert `data-model-bar` / `data-bar-current` attributes in
+`components/benchmark-detail.tsx` — keep them.
+
+## Known limitations (spot-check manually when touching the relevant code)
+
+- The chart check only inspects the DEFAULT-rendered metric tab; a break isolated to a
+  non-default tab isn't exercised.
+- Eval/benchmark detail pages get liveness-only checking — their leaderboard content isn't
+  asserted against the index the way model charts are.
+- Error-marker strings are hardcoded UI copy; update them here if the copy changes.
+- **Redirect preservation** (old/pre-rework URLs that used to redirect): not gated here. The
+  warehouse-derived map only covers ids the producer folds into `raw_model_ids`; old base→variant
+  spellings that aren't folded will 404. That's a stale-bookmark concern with an upstream
+  (registry alias) fix, not a frontend gate.
+
+## Enforcement
+
+There is no CI today (this repo is an HF Space — deploy is `git push`; HF builds the Dockerfile
+and does not run tests). So these are run by hand, by convention, like the `scripts/verify-*.mjs`
+family. If we add hosted CI later (e.g. mirror to GitHub), `test:integrity` is the cheap,
+server-free job to wire in first; `test:e2e` belongs in a separate, explicitly-owned job.

tests/e2e/frontend-preflight.spec.ts

@@ -0,0 +1,169 @@
+import { DuckDBConnection } from "@duckdb/node-api"
+import { expect, test, type Browser } from "@playwright/test"
+
+// Frontend correctness e2e — the layer naive "page returns 200" checks miss.
+// Opt-in via `SNAPSHOT_URL=<warehouse> pnpm test:e2e`; self-skips otherwise.
+// Verifies, against the live server + comparison-index ground truth:
+//  - model/eval/developer pages render (incl. folded-id + encoded-name regressions),
+//  - 100% of folded model ids resolve (the raw_model_ids fallback),
+//  - comparison charts render real PEER bars (not just the current model), 0 "Unknown Model".
+// Full bug taxonomy + known limitations: tests/PREFLIGHT.md.
+
+const BASE = `http://localhost:${process.env.PORT || 3211}`
+const SNAPSHOT = (process.env.SNAPSHOT_URL || "").replace(/\/+$/, "")
+const SAMPLE = 30
+const CHART_MODELS = 16
+
+test.describe.configure({ mode: "serial" })
+test.skip(!SNAPSHOT, "set SNAPSHOT_URL to run the frontend preflight e2e")
+
+const ERROR_MARKERS = [
+  "Model not found", "Failed to load model data", "Eval not found",
+  "Benchmark not found", "Failed to load", "Application error",
+  "Something went wrong", "This page could not be found",
+]
+
+const enc = (s: unknown) => encodeURIComponent(String(s))
+const norm = (s: unknown) => String(s || "").toLowerCase().replace(/[^a-z0-9]/g, "")
+const asArray = (j: any): any[] => Array.isArray(j) ? j
+  : (j && typeof j === "object" && ["models", "evals", "developers", "items", "rows", "cards", "data"].map((k) => j[k]).find(Array.isArray)) || []
+const sample = <T>(a: T[], n: number): T[] => a.length <= n ? a.slice() : Array.from({ length: n }, (_, i) => a[Math.floor(i * (a.length / n))])
+const listItems = (v: any): string[] => Array.isArray(v) ? v.map(String) : Array.isArray(v?.items) ? v.items.map(String) : []
+const getJson = async (p: string) => { const r = await fetch(`${BASE}${p}`); return r.ok ? r.json() : null }
+const getJsonT = async (p: string, ms = 90_000, tries = 2) => {
+  for (let i = 0; i < tries; i++) {
+    try { const r = await fetch(`${BASE}${p}`, { signal: AbortSignal.timeout(ms) }); if (r.ok) return r.json() } catch { /* retry */ }
+  }
+  return null
+}
+async function mapLimit<T, R>(items: T[], limit: number, fn: (x: T) => Promise<R>): Promise<R[]> {
+  const out = new Array<R>(items.length)
+  let i = 0
+  await Promise.all(Array.from({ length: Math.min(limit, items.length) }, async () => {
+    while (i < items.length) { const idx = i++; out[idx] = await fn(items[idx]) }
+  }))
+  return out
+}
+
+let models: any[] = []
+let evals: any[] = []
+let devs: any[] = []
+let folded: string[] = []
+let expectedPeersFor: (routeId: string) => string[]
+
+test.beforeAll(async () => {
+  models = asArray(await getJson("/api/model-cards-lite"))
+  evals = asArray(await getJson("/api/eval-list-lite"))
+  devs = asArray(await getJson("/api/developers"))
+  expect(models.length, "model-cards-lite empty").toBeGreaterThan(0)
+
+  // Comparison-index is the chart ground truth — unavailable must be a HARD fail,
+  // not a silent skip (else the whole chart layer disables itself).
+  const ci = await getJsonT("/api/comparison-index")
+  expect(ci?.by_model && ci?.evals, "comparison-index unavailable — chart check cannot run").toBeTruthy()
+  expectedPeersFor = (routeId: string) => {
+    const byModel = ci.by_model[routeId]
+    if (!byModel) return []
+    const peers = new Set<string>()
+    for (const evalId of Object.keys(byModel)) {
+      for (const metric of (ci.evals[evalId]?.metrics ?? [])) {
+        for (const s of metric.scores) {
+          if (s.model_route_id === routeId) continue
+          const n = norm(s.model_family_name || s.model_family_id)
+          if (n) peers.add(n)
+        }
+      }
+    }
+    return [...peers]
+  }
+
+  // Discover folded ids exhaustively from the warehouse (not a tiny stride).
+  const con = await DuckDBConnection.create()
+  await con.run("INSTALL httpfs; LOAD httpfs;")
+  const rows = (await con.runAndReadAll(
+    `SELECT model_id, raw_model_ids FROM read_parquet('${SNAPSHOT}/models_view.parquet') WHERE len(raw_model_ids) > 0`,
+  )).getRowObjects()
+  const set = new Set<string>()
+  for (const r of rows) for (const raw of listItems(r.raw_model_ids)) {
+    if (String(raw).toLowerCase() !== String(r.model_id).toLowerCase()) set.add(String(raw))
+  }
+  folded = [...set]
+  expect(folded.length, "folded-id regression set empty — discovery/data path changed").toBeGreaterThan(0)
+})
+
+test("100% of folded model ids resolve (raw_model_ids fallback)", async () => {
+  const statuses = await mapLimit(folded, 12, async (raw) => {
+    const r = await fetch(`${BASE}/api/model-summary?id=${enc(raw)}`, { signal: AbortSignal.timeout(30_000) }).catch(() => null)
+    return { raw, status: r ? r.status : 0 }
+  })
+  const bad = statuses.filter((s) => s.status !== 200).map((s) => s.raw)
+  expect(bad, `${bad.length}/${folded.length} folded ids do not resolve: ${bad.slice(0, 8)}`).toEqual([])
+})
+
+test("model / eval / developer pages render (incl. regression sets)", async ({ browser }) => {
+  const seen = new Set<string>()
+  const targets: { cls: string; url: string }[] = []
+  const add = (cls: string, url: string) => { if (url && !seen.has(url)) { seen.add(url); targets.push({ cls, url }) } }
+
+  for (const m of sample(models, SAMPLE)) add("model", `/models/${enc(m.model_id || m.id || m.model_key)}`)
+  for (const raw of sample(folded, 25)) add("model(folded)", `/models/${enc(raw)}`)
+  for (const e of sample(evals, SAMPLE)) { const id = e.evaluation_id || e.id || e.benchmark_id; if (id) add("eval", `/evals/${String(id).replace(/%2F/g, "/")}`) }
+  const devUrl = (rid: string) => `/developers/${String(rid).replace(/%2F/g, "/")}`
+  const encodedDevs = devs.filter((d) => /[^A-Za-z0-9._/-]/.test(String(d.developer || "")) || /%(?!2F)/i.test(String(d.route_id || "")))
+  expect(encodedDevs.length, "encoded-developer regression set empty — data path changed").toBeGreaterThan(0)
+  for (const d of encodedDevs) if (d.route_id) add("developer(encoded)", devUrl(d.route_id))
+  for (const d of sample(devs, 20)) if (d.route_id) add("developer", devUrl(d.route_id))
+
+  const results = await mapLimit(targets, 6, async (t) => {
+    const page = await browser.newPage()
+    const errs: string[] = []
+    page.on("console", (m) => { if (m.type() === "error") errs.push(m.text().slice(0, 120)) })
+    page.on("pageerror", (e) => errs.push(String(e).slice(0, 120)))
+    let status = 0
+    try { const r = await page.goto(`${BASE}${t.url}`, { waitUntil: "networkidle", timeout: 45_000 }); status = r ? r.status() : 0; await page.waitForTimeout(700) }
+    catch (e) { errs.push(`goto: ${String(e).slice(0, 120)}`) }
+    const text = await page.evaluate(() => document.body?.innerText || "").catch(() => "")
+    const marker = ERROR_MARKERS.find((m) => text.includes(m))
+    await page.close()
+    return (!(status > 0 && status < 400) || marker || errs.length)
+      ? `${t.url} [${status}]${marker ? ` "${marker}"` : ""}${errs.length ? ` ${JSON.stringify(errs.slice(0, 2))}` : ""}`
+      : null
+  })
+  const broken = results.filter(Boolean) as string[]
+  expect(broken, `${broken.length} broken pages:\n  ${broken.slice(0, 15).join("\n  ")}`).toEqual([])
+})
+
+test("comparison charts render real peer bars (not only the current model)", async ({ browser }) => {
+  const candidates = sample(models, CHART_MODELS)
+    .map((m) => enc(m.model_id || m.id || m.model_key))
+    .filter((routeId) => expectedPeersFor(routeId).length > 0)
+  const per = await mapLimit(candidates, 6, async (routeId) => {
+    const expected = expectedPeersFor(routeId)
+    const page = await browser.newPage()
+    const problems: string[] = []
+    let bars: { id: string | null; cur: boolean }[] = []
+    try {
+      await page.goto(`${BASE}/models/${routeId}`, { waitUntil: "networkidle", timeout: 45_000 })
+      await page.waitForTimeout(1000)
+      const unknown = ((await page.evaluate(() => document.body?.innerText || "")).match(/Unknown Model/g) || []).length
+      bars = await page.$$eval("[data-model-bar]", (els) =>
+        els.map((e) => ({ id: e.getAttribute("data-model-bar"), cur: e.getAttribute("data-bar-current") === "1" })))
+      const peerBars = bars.filter((b) => !b.cur)
+      const currentBars = bars.length - peerBars.length
+      // MULTIPLE charts rendered but NONE show a peer bar => the only-current-model bug.
+      if (currentBars >= 2 && peerBars.length === 0) problems.push(`${routeId}: ${expected.length} peers in index but 0 peer bars`)
+      if (peerBars.some((b) => b.id === routeId)) problems.push(`${routeId}: current model rendered as its own peer (double-count)`)
+      if (unknown > 0) problems.push(`${routeId}: ${unknown} "Unknown Model" labels`)
+    } catch (e) { problems.push(`${routeId}: ${String(e).slice(0, 80)}`) }
+    await page.close()
+    return { bars: bars.length, problems }
+  })
+  const charted = candidates.length
+  const totalBars = per.reduce((n, r) => n + r.bars, 0)
+  const broken = per.flatMap((r) => r.problems)
+  expect(charted, "no sampled model had expected peers — chart check was vacuous").toBeGreaterThan(0)
+  // Suite-wide: peers expected but NO bars anywhere => the data-model-bar hook was
+  // dropped or charts don't render — the chart check silently disabled itself.
+  expect(totalBars, "0 chart bars across charted pages — data-model-bar hook missing or charts broken").toBeGreaterThan(0)
+  expect(broken, `chart problems:\n  ${broken.slice(0, 15).join("\n  ")}`).toEqual([])
+})

tests/fixtures/developers/01-ai.json

@@ -2,7 +2,7 @@
   "developer": "01-ai",
   "models": [
     {
-      "model_family_id": "01-ai/yi-34b",
+      "model_group_id": "01-ai/yi-34b",
       "model_route_id": "01-ai__yi-34b",
       "model_family_name": "Yi 34B",
       "developer": "01-ai",
@@ -215,7 +215,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-6b",
+      "model_group_id": "01-ai/yi-6b",
       "model_route_id": "01-ai__yi-6b",
       "model_family_name": "Yi 6B",
       "developer": "01-ai",
@@ -428,7 +428,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-34b-chat",
+      "model_group_id": "01-ai/yi-34b-chat",
       "model_route_id": "01-ai__yi-34b-chat",
       "model_family_name": "Yi Chat 34B",
       "developer": "01-ai",
@@ -560,7 +560,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-large-preview",
+      "model_group_id": "01-ai/yi-large-preview",
       "model_route_id": "01-ai__yi-large-preview",
       "model_family_name": "Yi Large Preview",
       "developer": "01-ai",
@@ -730,7 +730,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-34b",
+      "model_group_id": "01-ai/yi-1-5-34b",
       "model_route_id": "01-ai__yi-1-5-34b",
       "model_family_name": "Yi-1.5-34B",
       "developer": "01-ai",
@@ -851,7 +851,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-34b-32k",
+      "model_group_id": "01-ai/yi-1-5-34b-32k",
       "model_route_id": "01-ai__yi-1-5-34b-32k",
       "model_family_name": "Yi-1.5-34B-32K",
       "developer": "01-ai",
@@ -972,7 +972,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-34b-chat",
+      "model_group_id": "01-ai/yi-1-5-34b-chat",
       "model_route_id": "01-ai__yi-1-5-34b-chat",
       "model_family_name": "Yi-1.5-34B-Chat",
       "developer": "01-ai",
@@ -1093,7 +1093,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-34b-chat-16k",
+      "model_group_id": "01-ai/yi-1-5-34b-chat-16k",
       "model_route_id": "01-ai__yi-1-5-34b-chat-16k",
       "model_family_name": "Yi-1.5-34B-Chat-16K",
       "developer": "01-ai",
@@ -1214,7 +1214,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-6b",
+      "model_group_id": "01-ai/yi-1-5-6b",
       "model_route_id": "01-ai__yi-1-5-6b",
       "model_family_name": "Yi-1.5-6B",
       "developer": "01-ai",
@@ -1335,7 +1335,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-6b-chat",
+      "model_group_id": "01-ai/yi-1-5-6b-chat",
       "model_route_id": "01-ai__yi-1-5-6b-chat",
       "model_family_name": "Yi-1.5-6B-Chat",
       "developer": "01-ai",
@@ -1456,7 +1456,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-9b",
+      "model_group_id": "01-ai/yi-1-5-9b",
       "model_route_id": "01-ai__yi-1-5-9b",
       "model_family_name": "Yi-1.5-9B",
       "developer": "01-ai",
@@ -1577,7 +1577,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-9b-32k",
+      "model_group_id": "01-ai/yi-1-5-9b-32k",
       "model_route_id": "01-ai__yi-1-5-9b-32k",
       "model_family_name": "Yi-1.5-9B-32K",
       "developer": "01-ai",
@@ -1698,7 +1698,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-9b-chat",
+      "model_group_id": "01-ai/yi-1-5-9b-chat",
       "model_route_id": "01-ai__yi-1-5-9b-chat",
       "model_family_name": "Yi-1.5-9B-Chat",
       "developer": "01-ai",
@@ -1819,7 +1819,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-1-5-9b-chat-16k",
+      "model_group_id": "01-ai/yi-1-5-9b-chat-16k",
       "model_route_id": "01-ai__yi-1-5-9b-chat-16k",
       "model_family_name": "Yi-1.5-9B-Chat-16K",
       "developer": "01-ai",
@@ -1940,7 +1940,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-34b-200k",
+      "model_group_id": "01-ai/yi-34b-200k",
       "model_route_id": "01-ai__yi-34b-200k",
       "model_family_name": "Yi-34B-200K",
       "developer": "01-ai",
@@ -2061,7 +2061,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-6b-200k",
+      "model_group_id": "01-ai/yi-6b-200k",
       "model_route_id": "01-ai__yi-6b-200k",
       "model_family_name": "Yi-6B-200K",
       "developer": "01-ai",
@@ -2182,7 +2182,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-6b-chat",
+      "model_group_id": "01-ai/yi-6b-chat",
       "model_route_id": "01-ai__yi-6b-chat",
       "model_family_name": "Yi-6B-Chat",
       "developer": "01-ai",
@@ -2303,7 +2303,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-9b",
+      "model_group_id": "01-ai/yi-9b",
       "model_route_id": "01-ai__yi-9b",
       "model_family_name": "Yi-9B",
       "developer": "01-ai",
@@ -2424,7 +2424,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-9b-200k",
+      "model_group_id": "01-ai/yi-9b-200k",
       "model_route_id": "01-ai__yi-9b-200k",
       "model_family_name": "Yi-9B-200K",
       "developer": "01-ai",
@@ -2545,7 +2545,7 @@
       ]
     },
     {
-      "model_family_id": "01-ai/yi-coder-9b-chat",
+      "model_group_id": "01-ai/yi-coder-9b-chat",
       "model_route_id": "01-ai__yi-coder-9b-chat",
       "model_family_name": "Yi-Coder-9B-Chat",
       "developer": "01-ai",

tests/fixtures/developers/anthropic.json

@@ -2,7 +2,7 @@
   "developer": "Anthropic",
   "models": [
     {
-      "model_family_id": "anthropic/claude-v1-3",
+      "model_group_id": "anthropic/claude-v1-3",
       "model_route_id": "anthropic__claude-v1-3",
       "model_family_name": "Anthropic Claude v1.3",
       "developer": "anthropic",
@@ -217,7 +217,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-2-0",
+      "model_group_id": "anthropic/claude-2-0",
       "model_route_id": "anthropic__claude-2-0",
       "model_family_name": "Claude 2.0",
       "developer": "anthropic",
@@ -378,7 +378,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-2-1",
+      "model_group_id": "anthropic/claude-2-1",
       "model_route_id": "anthropic__claude-2-1",
       "model_family_name": "Claude 2.1",
       "developer": "anthropic",
@@ -548,7 +548,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-21",
+      "model_group_id": "anthropic/claude-21",
       "model_route_id": "anthropic__claude-21",
       "model_family_name": "Claude 2.1",
       "developer": "anthropic",
@@ -743,7 +743,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-haiku",
+      "model_group_id": "anthropic/claude-3-haiku",
       "model_route_id": "anthropic__claude-3-haiku",
       "model_family_name": "Claude 3 Haiku",
       "developer": "anthropic",
@@ -978,7 +978,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-opus",
+      "model_group_id": "anthropic/claude-3-opus",
       "model_route_id": "anthropic__claude-3-opus",
       "model_family_name": "Claude 3 Opus",
       "developer": "anthropic",
@@ -1214,7 +1214,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-sonnet",
+      "model_group_id": "anthropic/claude-3-sonnet",
       "model_route_id": "anthropic__claude-3-sonnet",
       "model_family_name": "Claude 3 Sonnet",
       "developer": "anthropic",
@@ -1446,7 +1446,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-5-haiku",
+      "model_group_id": "anthropic/claude-3-5-haiku",
       "model_route_id": "anthropic__claude-3-5-haiku",
       "model_family_name": "Claude 3.5 Haiku",
       "developer": "anthropic",
@@ -1672,7 +1672,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-5-sonnet",
+      "model_group_id": "anthropic/claude-3-5-sonnet",
       "model_route_id": "anthropic__claude-3-5-sonnet",
       "model_family_name": "Claude 3.5 Sonnet",
       "developer": "anthropic",
@@ -1920,7 +1920,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-35-sonnet-june-24",
+      "model_group_id": "anthropic/claude-35-sonnet-june-24",
       "model_route_id": "anthropic__claude-35-sonnet-june-24",
       "model_family_name": "Claude 3.5 Sonnet (June '24)",
       "developer": "anthropic",
@@ -2106,7 +2106,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-35-sonnet",
+      "model_group_id": "anthropic/claude-35-sonnet",
       "model_route_id": "anthropic__claude-35-sonnet",
       "model_family_name": "Claude 3.5 Sonnet (Oct '24)",
       "developer": "anthropic",
@@ -2301,7 +2301,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7",
+      "model_group_id": "anthropic/claude-3-7",
       "model_route_id": "anthropic__claude-3-7",
       "model_family_name": "Claude 3.7",
       "developer": "anthropic",
@@ -2370,7 +2370,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7-sonnet",
+      "model_group_id": "anthropic/claude-3-7-sonnet",
       "model_route_id": "anthropic__claude-3-7-sonnet",
       "model_family_name": "Claude 3.7 Sonnet",
       "developer": "anthropic",
@@ -2613,7 +2613,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7-sonnet-thinking",
+      "model_group_id": "anthropic/claude-3-7-sonnet-thinking",
       "model_route_id": "anthropic__claude-3-7-sonnet-thinking",
       "model_family_name": "Claude 3.7 Sonnet (Reasoning)",
       "developer": "anthropic",
@@ -2808,7 +2808,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7-thinking-16k",
+      "model_group_id": "anthropic/claude-3-7-thinking-16k",
       "model_route_id": "anthropic__claude-3-7-thinking-16k",
       "model_family_name": "Claude 3.7 Thinking 16K",
       "developer": "anthropic",
@@ -2877,7 +2877,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7-thinking-1k",
+      "model_group_id": "anthropic/claude-3-7-thinking-1k",
       "model_route_id": "anthropic__claude-3-7-thinking-1k",
       "model_family_name": "Claude 3.7 Thinking 1K",
       "developer": "anthropic",
@@ -2946,7 +2946,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-7-thinking-8k",
+      "model_group_id": "anthropic/claude-3-7-thinking-8k",
       "model_route_id": "anthropic__claude-3-7-thinking-8k",
       "model_family_name": "Claude 3.7 Thinking 8K",
       "developer": "anthropic",
@@ -3015,7 +3015,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-opus-thinking",
+      "model_group_id": "anthropic/claude-4-opus-thinking",
       "model_route_id": "anthropic__claude-4-opus-thinking",
       "model_family_name": "Claude 4 Opus (Reasoning)",
       "developer": "anthropic",
@@ -3210,7 +3210,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-sonnet-thinking",
+      "model_group_id": "anthropic/claude-4-sonnet-thinking",
       "model_route_id": "anthropic__claude-4-sonnet-thinking",
       "model_family_name": "Claude 4 Sonnet (Reasoning)",
       "developer": "anthropic",
@@ -3405,7 +3405,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-1-opus",
+      "model_group_id": "anthropic/claude-4-1-opus",
       "model_route_id": "anthropic__claude-4-1-opus",
       "model_family_name": "Claude 4.1 Opus (Non-reasoning)",
       "developer": "anthropic",
@@ -3540,7 +3540,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-1-opus-thinking",
+      "model_group_id": "anthropic/claude-4-1-opus-thinking",
       "model_route_id": "anthropic__claude-4-1-opus-thinking",
       "model_family_name": "Claude 4.1 Opus (Reasoning)",
       "developer": "anthropic",
@@ -3735,7 +3735,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-5-haiku",
+      "model_group_id": "anthropic/claude-4-5-haiku",
       "model_route_id": "anthropic__claude-4-5-haiku",
       "model_family_name": "Claude 4.5 Haiku (Non-reasoning)",
       "developer": "anthropic",
@@ -3930,7 +3930,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-5-haiku-reasoning",
+      "model_group_id": "anthropic/claude-4-5-haiku-reasoning",
       "model_route_id": "anthropic__claude-4-5-haiku-reasoning",
       "model_family_name": "Claude 4.5 Haiku (Reasoning)",
       "developer": "anthropic",
@@ -4125,7 +4125,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-5-sonnet-thinking",
+      "model_group_id": "anthropic/claude-4-5-sonnet-thinking",
       "model_route_id": "anthropic__claude-4-5-sonnet-thinking",
       "model_family_name": "Claude 4.5 Sonnet (Reasoning)",
       "developer": "anthropic",
@@ -4320,7 +4320,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-code",
+      "model_group_id": "anthropic/claude-code",
       "model_route_id": "anthropic__claude-code",
       "model_family_name": "Claude Code",
       "developer": "Anthropic",
@@ -4389,7 +4389,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-instant",
+      "model_group_id": "anthropic/claude-instant",
       "model_route_id": "anthropic__claude-instant",
       "model_family_name": "Claude Instant",
       "developer": "anthropic",
@@ -4575,7 +4575,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-instant-1-2",
+      "model_group_id": "anthropic/claude-instant-1-2",
       "model_route_id": "anthropic__claude-instant-1-2",
       "model_family_name": "Claude Instant 1.2",
       "developer": "anthropic",
@@ -4745,7 +4745,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-mythos-preview",
+      "model_group_id": "anthropic/claude-mythos-preview",
       "model_route_id": "anthropic__claude-mythos-preview",
       "model_family_name": "Claude Mythos Preview",
       "developer": "anthropic",
@@ -4955,7 +4955,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4",
+      "model_group_id": "anthropic/claude-opus-4",
       "model_route_id": "anthropic__claude-opus-4",
       "model_family_name": "Claude Opus 4",
       "developer": "anthropic",
@@ -5220,7 +5220,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-5-thinking",
+      "model_group_id": "anthropic/claude-opus-4-5-thinking",
       "model_route_id": "anthropic__claude-opus-4-5-thinking",
       "model_family_name": "Claude Opus 4.5 (Reasoning)",
       "developer": "anthropic",
@@ -5415,7 +5415,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6-adaptive",
+      "model_group_id": "anthropic/claude-opus-4-6-adaptive",
       "model_route_id": "anthropic__claude-opus-4-6-adaptive",
       "model_family_name": "Claude Opus 4.6 (Adaptive Reasoning, Max Effort)",
       "developer": "anthropic",
@@ -5610,7 +5610,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4",
+      "model_group_id": "anthropic/claude-sonnet-4",
       "model_route_id": "anthropic__claude-sonnet-4",
       "model_family_name": "Claude Sonnet 4",
       "developer": "anthropic",
@@ -5875,7 +5875,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-6-adaptive",
+      "model_group_id": "anthropic/claude-sonnet-4-6-adaptive",
       "model_route_id": "anthropic__claude-sonnet-4-6-adaptive",
       "model_family_name": "Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)",
       "developer": "anthropic",
@@ -6070,7 +6070,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-6-non-reasoning-low-effort",
+      "model_group_id": "anthropic/claude-sonnet-4-6-non-reasoning-low-effort",
       "model_route_id": "anthropic__claude-sonnet-4-6-non-reasoning-low-effort",
       "model_family_name": "Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
       "developer": "anthropic",
@@ -6265,7 +6265,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-3-5-sonnet-oct",
+      "model_group_id": "anthropic/claude-3-5-sonnet-oct",
       "model_route_id": "anthropic__claude-3-5-sonnet-oct",
       "model_family_name": "Claude-3.5-Sonnet(Oct)",
       "developer": "anthropic",
@@ -6385,7 +6385,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-sonnet",
+      "model_group_id": "anthropic/claude-4-sonnet",
       "model_route_id": "anthropic__claude-4-sonnet",
       "model_family_name": "Claude-4-Sonnet",
       "developer": "anthropic",
@@ -6604,7 +6604,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-5-sonnet",
+      "model_group_id": "anthropic/claude-4-5-sonnet",
       "model_route_id": "anthropic__claude-4-5-sonnet",
       "model_family_name": "Claude-4.5-Sonnet",
       "developer": "anthropic",
@@ -6802,7 +6802,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-6-high",
+      "model_group_id": "anthropic/claude-sonnet-4-6-high",
       "model_route_id": "anthropic__claude-sonnet-4-6-high",
       "model_family_name": "Claude-Sonnet-4.6-high",
       "developer": "Anthropic",
@@ -6883,7 +6883,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/opus-4-1",
+      "model_group_id": "anthropic/opus-4-1",
       "model_route_id": "anthropic__opus-4-1",
       "model_family_name": "Opus 4.1",
       "developer": "anthropic",
@@ -6952,7 +6952,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/opus-4-5",
+      "model_group_id": "anthropic/opus-4-5",
       "model_route_id": "anthropic__opus-4-5",
       "model_family_name": "Opus 4.5",
       "developer": "anthropic",
@@ -7043,7 +7043,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/opus-4-6",
+      "model_group_id": "anthropic/opus-4-6",
       "model_route_id": "anthropic__opus-4-6",
       "model_family_name": "Opus 4.6",
       "developer": "anthropic",
@@ -7112,7 +7112,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/sonnet-4-5",
+      "model_group_id": "anthropic/sonnet-4-5",
       "model_route_id": "anthropic__sonnet-4-5",
       "model_family_name": "Sonnet 4.5",
       "developer": "anthropic",
@@ -7181,7 +7181,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/opus-4-6-max-effort",
+      "model_group_id": "anthropic/opus-4-6-max-effort",
       "model_route_id": "anthropic__opus-4-6-max-effort",
       "model_family_name": "anthropic-opus-4-6-max-effort",
       "developer": "anthropic",
@@ -7250,7 +7250,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/anthropic-claude-3-7-sonnet",
+      "model_group_id": "anthropic/anthropic-claude-3-7-sonnet",
       "model_route_id": "anthropic__anthropic-claude-3-7-sonnet",
       "model_family_name": "anthropic/anthropic/claude-3-7-sonnet-20250219",
       "developer": "anthropic",
@@ -7319,7 +7319,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-2",
+      "model_group_id": "anthropic/claude-2",
       "model_route_id": "anthropic__claude-2",
       "model_family_name": "claude-2",
       "developer": "anthropic",
@@ -7507,7 +7507,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-4-opus",
+      "model_group_id": "anthropic/claude-4-opus",
       "model_route_id": "anthropic__claude-4-opus",
       "model_family_name": "claude-4-opus-20250514",
       "developer": "anthropic",
@@ -7715,7 +7715,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-haiku-4-5",
+      "model_group_id": "anthropic/claude-haiku-4-5",
       "model_route_id": "anthropic__claude-haiku-4-5",
       "model_family_name": "claude-haiku-4-5-20251001",
       "developer": "anthropic",
@@ -7985,7 +7985,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-1",
+      "model_group_id": "anthropic/claude-opus-4-1",
       "model_route_id": "anthropic__claude-opus-4-1",
       "model_family_name": "claude-opus-4-1-20250805",
       "developer": "anthropic",
@@ -8125,7 +8125,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-5",
+      "model_group_id": "anthropic/claude-opus-4-5",
       "model_route_id": "anthropic__claude-opus-4-5",
       "model_family_name": "claude-opus-4-5-20251101",
       "developer": "anthropic",
@@ -8398,7 +8398,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6",
+      "model_group_id": "anthropic/claude-opus-4-6",
       "model_route_id": "anthropic__claude-opus-4-6",
       "model_family_name": "claude-opus-4-6",
       "developer": "anthropic",
@@ -8624,7 +8624,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6-thinking-120k-high",
+      "model_group_id": "anthropic/claude-opus-4-6-thinking-120k-high",
       "model_route_id": "anthropic__claude-opus-4-6-thinking-120k-high",
       "model_family_name": "claude-opus-4-6-thinking-120K-high",
       "developer": "anthropic",
@@ -8693,7 +8693,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6-thinking-120k-low",
+      "model_group_id": "anthropic/claude-opus-4-6-thinking-120k-low",
       "model_route_id": "anthropic__claude-opus-4-6-thinking-120k-low",
       "model_family_name": "claude-opus-4-6-thinking-120K-low",
       "developer": "anthropic",
@@ -8762,7 +8762,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6-thinking-120k-max",
+      "model_group_id": "anthropic/claude-opus-4-6-thinking-120k-max",
       "model_route_id": "anthropic__claude-opus-4-6-thinking-120k-max",
       "model_family_name": "claude-opus-4-6-thinking-120K-max",
       "developer": "anthropic",
@@ -8831,7 +8831,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-6-thinking-120k-medium",
+      "model_group_id": "anthropic/claude-opus-4-6-thinking-120k-medium",
       "model_route_id": "anthropic__claude-opus-4-6-thinking-120k-medium",
       "model_family_name": "claude-opus-4-6-thinking-120K-medium",
       "developer": "anthropic",
@@ -8900,7 +8900,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-opus-4-7",
+      "model_group_id": "anthropic/claude-opus-4-7",
       "model_route_id": "anthropic__claude-opus-4-7",
       "model_family_name": "claude-opus-4-7",
       "developer": "anthropic",
@@ -9079,7 +9079,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-5",
+      "model_group_id": "anthropic/claude-sonnet-4-5",
       "model_route_id": "anthropic__claude-sonnet-4-5",
       "model_family_name": "claude-sonnet-4-5-20250929",
       "developer": "anthropic",
@@ -9353,7 +9353,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-6",
+      "model_group_id": "anthropic/claude-sonnet-4-6",
       "model_route_id": "anthropic__claude-sonnet-4-6",
       "model_family_name": "claude-sonnet-4-6",
       "developer": "anthropic",
@@ -9563,7 +9563,7 @@
       ]
     },
     {
-      "model_family_id": "anthropic/claude-sonnet-4-6-max",
+      "model_group_id": "anthropic/claude-sonnet-4-6-max",
       "model_route_id": "anthropic__claude-sonnet-4-6-max",
       "model_family_name": "claude_sonnet_4_6_max",
       "developer": "anthropic",

tests/fixtures/developers/openai.json

@@ -2,7 +2,7 @@
   "developer": "OpenAI",
   "models": [
     {
-      "model_family_id": "openai/chatgpt-agent",
+      "model_group_id": "openai/chatgpt-agent",
       "model_route_id": "openai__chatgpt-agent",
       "model_family_name": "ChatGPT Agent",
       "developer": "OpenAI",
@@ -71,7 +71,7 @@
       ]
     },
     {
-      "model_family_id": "openai/codex",
+      "model_group_id": "openai/codex",
       "model_route_id": "openai__codex",
       "model_family_name": "CodeX",
       "developer": "OpenAI",
@@ -140,7 +140,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-20b-high",
+      "model_group_id": "openai/gpt-oss-20b-high",
       "model_route_id": "openai__gpt-oss-20b-high",
       "model_family_name": "GPT OSS 20B High",
       "developer": "openai",
@@ -209,7 +209,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-3-5-turbo-0125",
+      "model_group_id": "openai/gpt-3-5-turbo-0125",
       "model_route_id": "openai__gpt-3-5-turbo-0125",
       "model_family_name": "GPT-3.5 Turbo",
       "developer": "openai",
@@ -380,7 +380,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-35-turbo",
+      "model_group_id": "openai/gpt-35-turbo",
       "model_route_id": "openai__gpt-35-turbo",
       "model_family_name": "GPT-3.5 Turbo",
       "developer": "openai",
@@ -539,7 +539,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-3-5-turbo-0301",
+      "model_group_id": "openai/gpt-3-5-turbo-0301",
       "model_route_id": "openai__gpt-3-5-turbo-0301",
       "model_family_name": "GPT-3.5 Turbo 0301",
       "developer": "openai",
@@ -753,7 +753,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-3-5-turbo-0613",
+      "model_group_id": "openai/gpt-3-5-turbo-0613",
       "model_route_id": "openai__gpt-3-5-turbo-0613",
       "model_family_name": "GPT-3.5 Turbo 0613",
       "developer": "openai",
@@ -990,7 +990,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-3-5-turbo-1106",
+      "model_group_id": "openai/gpt-3-5-turbo-1106",
       "model_route_id": "openai__gpt-3-5-turbo-1106",
       "model_family_name": "GPT-3.5 Turbo 1106",
       "developer": "openai",
@@ -1119,7 +1119,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-0613",
+      "model_group_id": "openai/gpt-4-0613",
       "model_route_id": "openai__gpt-4-0613",
       "model_family_name": "GPT-4",
       "developer": "openai",
@@ -1334,7 +1334,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4",
+      "model_group_id": "openai/gpt-4",
       "model_route_id": "openai__gpt-4",
       "model_family_name": "GPT-4",
       "developer": "openai",
@@ -1466,7 +1466,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-0314",
+      "model_group_id": "openai/gpt-4-0314",
       "model_route_id": "openai__gpt-4-0314",
       "model_family_name": "GPT-4 0314",
       "developer": "openai",
@@ -1598,7 +1598,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-turbo",
+      "model_group_id": "openai/gpt-4-turbo",
       "model_route_id": "openai__gpt-4-turbo",
       "model_family_name": "GPT-4 Turbo",
       "developer": "openai",
@@ -1825,7 +1825,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-1-nano",
+      "model_group_id": "openai/gpt-4-1-nano",
       "model_route_id": "openai__gpt-4-1-nano",
       "model_family_name": "GPT-4.1 nano",
       "developer": "openai",
@@ -2064,7 +2064,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-5",
+      "model_group_id": "openai/gpt-4-5",
       "model_route_id": "openai__gpt-4-5",
       "model_family_name": "GPT-4.5",
       "developer": "openai",
@@ -2290,7 +2290,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-5-preview",
+      "model_group_id": "openai/gpt-4-5-preview",
       "model_route_id": "openai__gpt-4-5-preview",
       "model_family_name": "GPT-4.5 2025-02-27 preview",
       "developer": "openai",
@@ -2419,7 +2419,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o",
+      "model_group_id": "openai/gpt-4o",
       "model_route_id": "openai__gpt-4o",
       "model_family_name": "GPT-4o",
       "developer": "openai",
@@ -2709,7 +2709,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-chatgpt",
+      "model_group_id": "openai/gpt-4o-chatgpt",
       "model_route_id": "openai__gpt-4o-chatgpt",
       "model_family_name": "GPT-4o (ChatGPT)",
       "developer": "openai",
@@ -2895,7 +2895,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-chatgpt-03-25",
+      "model_group_id": "openai/gpt-4o-chatgpt-03-25",
       "model_route_id": "openai__gpt-4o-chatgpt-03-25",
       "model_family_name": "GPT-4o (March 2025, chatgpt-4o-latest)",
       "developer": "openai",
@@ -3090,7 +3090,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-realtime-dec-2024",
+      "model_group_id": "openai/gpt-4o-realtime-dec-2024",
       "model_route_id": "openai__gpt-4o-realtime-dec-2024",
       "model_family_name": "GPT-4o Realtime (Dec '24)",
       "developer": "openai",
@@ -3204,7 +3204,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-mini",
+      "model_group_id": "openai/gpt-4o-mini",
       "model_route_id": "openai__gpt-4o-mini",
       "model_family_name": "GPT-4o mini",
       "developer": "openai",
@@ -3444,7 +3444,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-mini-realtime-dec-2024",
+      "model_group_id": "openai/gpt-4o-mini-realtime-dec-2024",
       "model_route_id": "openai__gpt-4o-mini-realtime-dec-2024",
       "model_family_name": "GPT-4o mini Realtime (Dec '24)",
       "developer": "openai",
@@ -3558,7 +3558,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4o-1120",
+      "model_group_id": "openai/gpt-4o-1120",
       "model_route_id": "openai__gpt-4o-1120",
       "model_family_name": "GPT-4o-1120",
       "developer": "openai",
@@ -3678,7 +3678,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5",
+      "model_group_id": "openai/gpt-5",
       "model_route_id": "openai__gpt-5",
       "model_family_name": "GPT-5",
       "developer": "openai",
@@ -3985,7 +3985,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-chatgpt",
+      "model_group_id": "openai/gpt-5-chatgpt",
       "model_route_id": "openai__gpt-5-chatgpt",
       "model_family_name": "GPT-5 (ChatGPT)",
       "developer": "openai",
@@ -4180,7 +4180,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-low",
+      "model_group_id": "openai/gpt-5-low",
       "model_route_id": "openai__gpt-5-low",
       "model_family_name": "GPT-5 (low)",
       "developer": "openai",
@@ -4375,7 +4375,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-medium",
+      "model_group_id": "openai/gpt-5-medium",
       "model_route_id": "openai__gpt-5-medium",
       "model_family_name": "GPT-5 (medium)",
       "developer": "openai",
@@ -4570,7 +4570,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-minimal",
+      "model_group_id": "openai/gpt-5-minimal",
       "model_route_id": "openai__gpt-5-minimal",
       "model_family_name": "GPT-5 (minimal)",
       "developer": "openai",
@@ -4765,7 +4765,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-codex",
+      "model_group_id": "openai/gpt-5-codex",
       "model_route_id": "openai__gpt-5-codex",
       "model_family_name": "GPT-5 Codex (high)",
       "developer": "openai",
@@ -4964,7 +4964,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-mini-medium",
+      "model_group_id": "openai/gpt-5-mini-medium",
       "model_route_id": "openai__gpt-5-mini-medium",
       "model_family_name": "GPT-5 mini (medium)",
       "developer": "openai",
@@ -5159,7 +5159,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-mini-minimal",
+      "model_group_id": "openai/gpt-5-mini-minimal",
       "model_route_id": "openai__gpt-5-mini-minimal",
       "model_family_name": "GPT-5 mini (minimal)",
       "developer": "openai",
@@ -5354,7 +5354,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-nano",
+      "model_group_id": "openai/gpt-5-nano",
       "model_route_id": "openai__gpt-5-nano",
       "model_family_name": "GPT-5 nano",
       "developer": "openai",
@@ -5618,7 +5618,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-nano-medium",
+      "model_group_id": "openai/gpt-5-nano-medium",
       "model_route_id": "openai__gpt-5-nano-medium",
       "model_family_name": "GPT-5 nano (medium)",
       "developer": "openai",
@@ -5813,7 +5813,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-nano-minimal",
+      "model_group_id": "openai/gpt-5-nano-minimal",
       "model_route_id": "openai__gpt-5-nano-minimal",
       "model_family_name": "GPT-5 nano (minimal)",
       "developer": "openai",
@@ -6008,7 +6008,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-non-reasoning",
+      "model_group_id": "openai/gpt-5-1-non-reasoning",
       "model_route_id": "openai__gpt-5-1-non-reasoning",
       "model_family_name": "GPT-5.1 (Non-reasoning)",
       "developer": "openai",
@@ -6203,7 +6203,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-codex-mini",
+      "model_group_id": "openai/gpt-5-1-codex-mini",
       "model_route_id": "openai__gpt-5-1-codex-mini",
       "model_family_name": "GPT-5.1 Codex mini (high)",
       "developer": "openai",
@@ -6401,7 +6401,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-high",
+      "model_group_id": "openai/gpt-5-1-high",
       "model_route_id": "openai__gpt-5-1-high",
       "model_family_name": "GPT-5.1 High",
       "developer": "openai",
@@ -6470,7 +6470,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-codex-max",
+      "model_group_id": "openai/gpt-5-1-codex-max",
       "model_route_id": "openai__gpt-5-1-codex-max",
       "model_family_name": "GPT-5.1-Codex-Max",
       "developer": "OpenAI",
@@ -6539,7 +6539,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-2-non-reasoning",
+      "model_group_id": "openai/gpt-5-2-non-reasoning",
       "model_route_id": "openai__gpt-5-2-non-reasoning",
       "model_family_name": "GPT-5.2 (Non-reasoning)",
       "developer": "openai",
@@ -6734,7 +6734,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-2-medium",
+      "model_group_id": "openai/gpt-5-2-medium",
       "model_route_id": "openai__gpt-5-2-medium",
       "model_family_name": "GPT-5.2 (medium)",
       "developer": "openai",
@@ -6929,7 +6929,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-2-pro",
+      "model_group_id": "openai/gpt-5-2-pro",
       "model_route_id": "openai__gpt-5-2-pro",
       "model_family_name": "GPT-5.2 Pro",
       "developer": "openai",
@@ -7106,7 +7106,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-non-reasoning",
+      "model_group_id": "openai/gpt-5-4-non-reasoning",
       "model_route_id": "openai__gpt-5-4-non-reasoning",
       "model_family_name": "GPT-5.4 (Non-reasoning)",
       "developer": "openai",
@@ -7301,7 +7301,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-pro",
+      "model_group_id": "openai/gpt-5-4-pro",
       "model_route_id": "openai__gpt-5-4-pro",
       "model_family_name": "GPT-5.4 Pro (xhigh)",
       "developer": "openai",
@@ -7415,7 +7415,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini-non-reasoning",
+      "model_group_id": "openai/gpt-5-4-mini-non-reasoning",
       "model_route_id": "openai__gpt-5-4-mini-non-reasoning",
       "model_family_name": "GPT-5.4 mini (Non-Reasoning)",
       "developer": "openai",
@@ -7610,7 +7610,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini-medium",
+      "model_group_id": "openai/gpt-5-4-mini-medium",
       "model_route_id": "openai__gpt-5-4-mini-medium",
       "model_family_name": "GPT-5.4 mini (medium)",
       "developer": "openai",
@@ -7807,7 +7807,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano-non-reasoning",
+      "model_group_id": "openai/gpt-5-4-nano-non-reasoning",
       "model_route_id": "openai__gpt-5-4-nano-non-reasoning",
       "model_family_name": "GPT-5.4 nano (Non-Reasoning)",
       "developer": "openai",
@@ -8002,7 +8002,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano-medium",
+      "model_group_id": "openai/gpt-5-4-nano-medium",
       "model_route_id": "openai__gpt-5-4-nano-medium",
       "model_family_name": "GPT-5.4 nano (medium)",
       "developer": "openai",
@@ -8199,7 +8199,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-high",
+      "model_group_id": "openai/gpt-5-4-high",
       "model_route_id": "openai__gpt-5-4-high",
       "model_family_name": "GPT-5.4-high",
       "developer": "OpenAI",
@@ -8280,7 +8280,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-5",
+      "model_group_id": "openai/gpt-5-5",
       "model_route_id": "openai__gpt-5-5",
       "model_family_name": "GPT-5.5",
       "developer": "openai",
@@ -8497,7 +8497,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-5-pro",
+      "model_group_id": "openai/gpt-5-5-pro",
       "model_route_id": "openai__gpt-5-5-pro",
       "model_family_name": "GPT-5.5 Pro",
       "developer": "openai",
@@ -8606,7 +8606,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-j-6b",
+      "model_group_id": "openai/gpt-j-6b",
       "model_route_id": "openai__gpt-j-6b",
       "model_family_name": "GPT-J 6B",
       "developer": "openai",
@@ -8818,7 +8818,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-neox-20b",
+      "model_group_id": "openai/gpt-neox-20b",
       "model_route_id": "openai__gpt-neox-20b",
       "model_family_name": "GPT-NeoX 20B",
       "developer": "openai",
@@ -9030,7 +9030,7 @@
       ]
     },
     {
-      "model_family_id": "openai/lingma-agent-lingma-swe-gpt-72b-v0918",
+      "model_group_id": "openai/lingma-agent-lingma-swe-gpt-72b-v0918",
       "model_route_id": "openai__lingma-agent-lingma-swe-gpt-72b-v0918",
       "model_family_name": "Lingma Agent + Lingma SWE-GPT 72b (v0918)",
       "developer": "openai",
@@ -9099,7 +9099,7 @@
       ]
     },
     {
-      "model_family_id": "openai/lingma-agent-lingma-swe-gpt-72b-v0925",
+      "model_group_id": "openai/lingma-agent-lingma-swe-gpt-72b-v0925",
       "model_route_id": "openai__lingma-agent-lingma-swe-gpt-72b-v0925",
       "model_family_name": "Lingma Agent + Lingma SWE-GPT 72b (v0925)",
       "developer": "openai",
@@ -9168,7 +9168,7 @@
       ]
     },
     {
-      "model_family_id": "openai/lingma-agent-lingma-swe-gpt-7b-v0918",
+      "model_group_id": "openai/lingma-agent-lingma-swe-gpt-7b-v0918",
       "model_route_id": "openai__lingma-agent-lingma-swe-gpt-7b-v0918",
       "model_family_name": "Lingma Agent + Lingma SWE-GPT 7b (v0918)",
       "developer": "openai",
@@ -9237,7 +9237,7 @@
       ]
     },
     {
-      "model_family_id": "openai/lingma-agent-lingma-swe-gpt-7b-v0925",
+      "model_group_id": "openai/lingma-agent-lingma-swe-gpt-7b-v0925",
       "model_route_id": "openai__lingma-agent-lingma-swe-gpt-7b-v0925",
       "model_family_name": "Lingma Agent + Lingma SWE-GPT 7b (v0925)",
       "developer": "openai",
@@ -9306,7 +9306,7 @@
       ]
     },
     {
-      "model_family_id": "openai/deep-research",
+      "model_group_id": "openai/deep-research",
       "model_route_id": "openai__deep-research",
       "model_family_name": "OpenAI Deep Research",
       "developer": "OpenAI",
@@ -9375,7 +9375,7 @@
       ]
     },
     {
-      "model_family_id": "openai/openai-o1",
+      "model_group_id": "openai/openai-o1",
       "model_route_id": "openai__openai-o1",
       "model_family_name": "OpenAI-o1",
       "developer": "openai",
@@ -9495,7 +9495,7 @@
       ]
     },
     {
-      "model_family_id": "openai/openai-o3-mini-high",
+      "model_group_id": "openai/openai-o3-mini-high",
       "model_route_id": "openai__openai-o3-mini-high",
       "model_family_name": "OpenAI-o3-mini-high",
       "developer": "openai",
@@ -9615,7 +9615,7 @@
       ]
     },
     {
-      "model_family_id": "openai/ada-350m",
+      "model_group_id": "openai/ada-350m",
       "model_route_id": "openai__ada-350m",
       "model_family_name": "ada 350M",
       "developer": "openai",
@@ -9827,7 +9827,7 @@
       ]
     },
     {
-      "model_family_id": "openai/babbage-1-3b",
+      "model_group_id": "openai/babbage-1-3b",
       "model_route_id": "openai__babbage-1-3b",
       "model_family_name": "babbage 1.3B",
       "developer": "openai",
@@ -10039,7 +10039,7 @@
       ]
     },
     {
-      "model_family_id": "openai/codex-mini-latest",
+      "model_group_id": "openai/codex-mini-latest",
       "model_route_id": "openai__codex-mini-latest",
       "model_family_name": "codex-mini-latest",
       "developer": "openai",
@@ -10108,7 +10108,7 @@
       ]
     },
     {
-      "model_family_id": "openai/curie-6-7b",
+      "model_group_id": "openai/curie-6-7b",
       "model_route_id": "openai__curie-6-7b",
       "model_family_name": "curie 6.7B",
       "developer": "openai",
@@ -10320,7 +10320,7 @@
       ]
     },
     {
-      "model_family_id": "openai/davinci-175b",
+      "model_group_id": "openai/davinci-175b",
       "model_route_id": "openai__davinci-175b",
       "model_family_name": "davinci 175B",
       "developer": "openai",
@@ -10532,7 +10532,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-1106-preview",
+      "model_group_id": "openai/gpt-4-1106-preview",
       "model_route_id": "openai__gpt-4-1106-preview",
       "model_family_name": "gpt-4-1106-preview",
       "developer": "openai",
@@ -10713,7 +10713,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-1",
+      "model_group_id": "openai/gpt-4-1",
       "model_route_id": "openai__gpt-4-1",
       "model_family_name": "gpt-4.1-2025-04-14",
       "developer": "openai",
@@ -10967,7 +10967,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-1-mini",
+      "model_group_id": "openai/gpt-4-1-mini",
       "model_route_id": "openai__gpt-4-1-mini",
       "model_family_name": "gpt-4.1-mini-2025-04-14",
       "developer": "openai",
@@ -11211,7 +11211,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-low",
+      "model_group_id": "openai/gpt-5-4-low",
       "model_route_id": "openai__gpt-5-4-low",
       "model_family_name": "gpt-5-4-low",
       "developer": "openai",
@@ -11280,7 +11280,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-medium",
+      "model_group_id": "openai/gpt-5-4-medium",
       "model_route_id": "openai__gpt-5-4-medium",
       "model_family_name": "gpt-5-4-medium",
       "developer": "openai",
@@ -11349,7 +11349,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini-high",
+      "model_group_id": "openai/gpt-5-4-mini-high",
       "model_route_id": "openai__gpt-5-4-mini-high",
       "model_family_name": "gpt-5-4-mini-high",
       "developer": "openai",
@@ -11418,7 +11418,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini-low",
+      "model_group_id": "openai/gpt-5-4-mini-low",
       "model_route_id": "openai__gpt-5-4-mini-low",
       "model_family_name": "gpt-5-4-mini-low",
       "developer": "openai",
@@ -11487,7 +11487,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini-xhigh",
+      "model_group_id": "openai/gpt-5-4-mini-xhigh",
       "model_route_id": "openai__gpt-5-4-mini-xhigh",
       "model_family_name": "gpt-5-4-mini-xhigh",
       "developer": "openai",
@@ -11556,7 +11556,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano-high",
+      "model_group_id": "openai/gpt-5-4-nano-high",
       "model_route_id": "openai__gpt-5-4-nano-high",
       "model_family_name": "gpt-5-4-nano-high",
       "developer": "openai",
@@ -11625,7 +11625,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano-low",
+      "model_group_id": "openai/gpt-5-4-nano-low",
       "model_route_id": "openai__gpt-5-4-nano-low",
       "model_family_name": "gpt-5-4-nano-low",
       "developer": "openai",
@@ -11694,7 +11694,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano-xhigh",
+      "model_group_id": "openai/gpt-5-4-nano-xhigh",
       "model_route_id": "openai__gpt-5-4-nano-xhigh",
       "model_family_name": "gpt-5-4-nano-xhigh",
       "developer": "openai",
@@ -11763,7 +11763,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-pro-xhigh",
+      "model_group_id": "openai/gpt-5-4-pro-xhigh",
       "model_route_id": "openai__gpt-5-4-pro-xhigh",
       "model_family_name": "gpt-5-4-pro-xhigh",
       "developer": "openai",
@@ -11832,7 +11832,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-xhigh",
+      "model_group_id": "openai/gpt-5-4-xhigh",
       "model_route_id": "openai__gpt-5-4-xhigh",
       "model_family_name": "gpt-5-4-xhigh",
       "developer": "openai",
@@ -11901,7 +11901,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-mini",
+      "model_group_id": "openai/gpt-5-mini",
       "model_route_id": "openai__gpt-5-mini",
       "model_family_name": "gpt-5-mini-2025-08-07",
       "developer": "openai",
@@ -12168,7 +12168,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-pro",
+      "model_group_id": "openai/gpt-5-pro",
       "model_route_id": "openai__gpt-5-pro",
       "model_family_name": "gpt-5-pro-2025-10-06",
       "developer": "openai",
@@ -12237,7 +12237,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1",
+      "model_group_id": "openai/gpt-5-1",
       "model_route_id": "openai__gpt-5-1",
       "model_family_name": "gpt-5.1-2025-11-13",
       "developer": "openai",
@@ -12508,7 +12508,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-codex",
+      "model_group_id": "openai/gpt-5-1-codex",
       "model_route_id": "openai__gpt-5-1-codex",
       "model_family_name": "gpt-5.1-codex",
       "developer": "openai",
@@ -12709,7 +12709,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-codex-high",
+      "model_group_id": "openai/gpt-5-1-codex-high",
       "model_route_id": "openai__gpt-5-1-codex-high",
       "model_family_name": "gpt-5.1-codex-high",
       "developer": "openai",
@@ -12778,7 +12778,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-instant",
+      "model_group_id": "openai/gpt-5-1-instant",
       "model_route_id": "openai__gpt-5-1-instant",
       "model_family_name": "gpt-5.1-instant-2025-11-12",
       "developer": "openai",
@@ -12897,7 +12897,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-medium",
+      "model_group_id": "openai/gpt-5-1-medium",
       "model_route_id": "openai__gpt-5-1-medium",
       "model_family_name": "gpt-5.1-medium-2025-11-12",
       "developer": "openai",
@@ -12966,7 +12966,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-1-thinking",
+      "model_group_id": "openai/gpt-5-1-thinking",
       "model_route_id": "openai__gpt-5-1-thinking",
       "model_family_name": "gpt-5.1-thinking-2025-11-12",
       "developer": "openai",
@@ -13085,7 +13085,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-2",
+      "model_group_id": "openai/gpt-5-2",
       "model_route_id": "openai__gpt-5-2",
       "model_family_name": "gpt-5.2-2025-12-11",
       "developer": "openai",
@@ -13378,7 +13378,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-2-codex",
+      "model_group_id": "openai/gpt-5-2-codex",
       "model_route_id": "openai__gpt-5-2-codex",
       "model_family_name": "gpt-5.2-codex",
       "developer": "openai",
@@ -13580,7 +13580,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-3-chat-latest",
+      "model_group_id": "openai/gpt-5-3-chat-latest",
       "model_route_id": "openai__gpt-5-3-chat-latest",
       "model_family_name": "gpt-5.3-chat-latest",
       "developer": "openai",
@@ -13659,7 +13659,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-3-codex",
+      "model_group_id": "openai/gpt-5-3-codex",
       "model_route_id": "openai__gpt-5-3-codex",
       "model_family_name": "gpt-5.3-codex",
       "developer": "openai",
@@ -13864,7 +13864,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4",
+      "model_group_id": "openai/gpt-5-4",
       "model_route_id": "openai__gpt-5-4",
       "model_family_name": "gpt-5.4",
       "developer": "openai",
@@ -14089,7 +14089,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-mini",
+      "model_group_id": "openai/gpt-5-4-mini",
       "model_route_id": "openai__gpt-5-4-mini",
       "model_family_name": "gpt-5.4-mini",
       "developer": "openai",
@@ -14296,7 +14296,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-5-4-nano",
+      "model_group_id": "openai/gpt-5-4-nano",
       "model_route_id": "openai__gpt-5-4-nano",
       "model_family_name": "gpt-5.4-nano",
       "developer": "openai",
@@ -14501,7 +14501,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-120b-low",
+      "model_group_id": "openai/gpt-oss-120b-low",
       "model_route_id": "openai__gpt-oss-120b-low",
       "model_family_name": "gpt-oss-120B (low)",
       "developer": "openai",
@@ -14696,7 +14696,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-120b",
+      "model_group_id": "openai/gpt-oss-120b",
       "model_route_id": "openai__gpt-oss-120b",
       "model_family_name": "gpt-oss-120b",
       "developer": "openai",
@@ -14921,7 +14921,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-120b-high",
+      "model_group_id": "openai/gpt-oss-120b-high",
       "model_route_id": "openai__gpt-oss-120b-high",
       "model_family_name": "gpt-oss-120b-high",
       "developer": "openai",
@@ -15011,7 +15011,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-20b-low",
+      "model_group_id": "openai/gpt-oss-20b-low",
       "model_route_id": "openai__gpt-oss-20b-low",
       "model_family_name": "gpt-oss-20B (low)",
       "developer": "openai",
@@ -15206,7 +15206,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-oss-20b",
+      "model_group_id": "openai/gpt-oss-20b",
       "model_route_id": "openai__gpt-oss-20b",
       "model_family_name": "gpt-oss-20b",
       "developer": "openai",
@@ -15427,7 +15427,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o1",
+      "model_group_id": "openai/o1",
       "model_route_id": "openai__o1",
       "model_family_name": "o1",
       "developer": "openai",
@@ -15655,7 +15655,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o1-mini",
+      "model_group_id": "openai/o1-mini",
       "model_route_id": "openai__o1-mini",
       "model_family_name": "o1-mini",
       "developer": "openai",
@@ -15872,7 +15872,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o1-preview",
+      "model_group_id": "openai/o1-preview",
       "model_route_id": "openai__o1-preview",
       "model_family_name": "o1-preview",
       "developer": "openai",
@@ -16075,7 +16075,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o1-pro",
+      "model_group_id": "openai/o1-pro",
       "model_route_id": "openai__o1-pro",
       "model_family_name": "o1-pro",
       "developer": "openai",
@@ -16208,7 +16208,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o3",
+      "model_group_id": "openai/o3",
       "model_route_id": "openai__o3",
       "model_family_name": "o3",
       "developer": "openai",
@@ -16482,7 +16482,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o3-mini",
+      "model_group_id": "openai/o3-mini",
       "model_route_id": "openai__o3-mini",
       "model_family_name": "o3-mini",
       "developer": "openai",
@@ -16746,7 +16746,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o3-mini-high",
+      "model_group_id": "openai/o3-mini-high",
       "model_route_id": "openai__o3-mini-high",
       "model_family_name": "o3-mini (high)",
       "developer": "openai",
@@ -16941,7 +16941,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o3-pro",
+      "model_group_id": "openai/o3-pro",
       "model_route_id": "openai__o3-pro",
       "model_family_name": "o3-pro",
       "developer": "openai",
@@ -17122,7 +17122,7 @@
       ]
     },
     {
-      "model_family_id": "openai/o4-mini",
+      "model_group_id": "openai/o4-mini",
       "model_route_id": "openai__o4-mini",
       "model_family_name": "o4-mini",
       "developer": "openai",
@@ -17390,7 +17390,7 @@
       ]
     },
     {
-      "model_family_id": "openai/gpt-4-0125-preview",
+      "model_group_id": "openai/gpt-4-0125-preview",
       "model_route_id": "openai__gpt-4-0125-preview",
       "model_family_name": "openai/gpt-4-0125-preview",
       "developer": "openai",
@@ -17468,7 +17468,7 @@
       ]
     },
     {
-      "model_family_id": "openai/text-ada-001",
+      "model_group_id": "openai/text-ada-001",
       "model_route_id": "openai__text-ada-001",
       "model_family_name": "text-ada-001",
       "developer": "openai",
@@ -17680,7 +17680,7 @@
       ]
     },
     {
-      "model_family_id": "openai/text-babbage-001",
+      "model_group_id": "openai/text-babbage-001",
       "model_route_id": "openai__text-babbage-001",
       "model_family_name": "text-babbage-001",
       "developer": "openai",
@@ -17892,7 +17892,7 @@
       ]
     },
     {
-      "model_family_id": "openai/text-curie-001",
+      "model_group_id": "openai/text-curie-001",
       "model_route_id": "openai__text-curie-001",
       "model_family_name": "text-curie-001",
       "developer": "openai",
@@ -18104,7 +18104,7 @@
       ]
     },
     {
-      "model_family_id": "openai/text-davinci-002",
+      "model_group_id": "openai/text-davinci-002",
       "model_route_id": "openai__text-davinci-002",
       "model_family_name": "text-davinci-002",
       "developer": "openai",
@@ -18324,7 +18324,7 @@
       ]
     },
     {
-      "model_family_id": "openai/text-davinci-003",
+      "model_group_id": "openai/text-davinci-003",
       "model_route_id": "openai__text-davinci-003",
       "model_family_name": "text-davinci-003",
       "developer": "openai",

tests/fixtures/model-cards/01-ai__yi-34b.json

@@ -1,5 +1,5 @@
 {
-  "model_family_id": "01-ai/yi-34b",
+  "model_group_id": "01-ai/yi-34b",
   "model_route_id": "01-ai__yi-34b",
   "model_family_name": "Yi 34B",
   "developer": "01-ai",

tests/fixtures/model-cards/anthropic__claude-opus-4.5.json

@@ -1,5 +1,5 @@
 {
-  "model_family_id": "anthropic/claude-opus-4.5",
+  "model_group_id": "anthropic/claude-opus-4.5",
   "model_route_id": "anthropic__claude-opus-4.5",
   "model_family_name": "Claude Opus 4.5",
   "developer": "anthropic",

tests/fixtures/model-cards/openai__gpt-5.json

@@ -1,5 +1,5 @@
 {
-  "model_family_id": "openai/gpt-5",
+  "model_group_id": "openai/gpt-5",
   "model_route_id": "openai__gpt-5",
   "model_family_name": "GPT 5",
   "developer": "openai",

tests/fixtures/models/ai21__j1-grande-v1-17b.json

@@ -13,7 +13,7 @@
     "model_route_id": "ai21__j1-grande-v1-17b",
     "model_version": null
   },
-  "model_family_id": "ai21/j1-grande-v1-17b",
+  "model_group_id": "ai21/j1-grande-v1-17b",
   "model_route_id": "ai21__j1-grande-v1-17b",
   "model_family_name": "J1-Grande v1 17B",
   "raw_model_ids": [

tests/fixtures/models/bytedance__seed-2-0-lite.json

@@ -25,7 +25,7 @@
     "model_route_id": "bytedance__seed-2-0-lite",
     "model_version": null
   },
-  "model_family_id": "bytedance/seed-2-0-lite",
+  "model_group_id": "bytedance/seed-2-0-lite",
   "model_route_id": "bytedance__seed-2-0-lite",
   "model_family_name": "Seed 2.0 Lite",
   "raw_model_ids": [

tests/fixtures/models/google__gemini-3-flash.json

@@ -22,7 +22,7 @@
     "model_route_id": "google__gemini-3-flash",
     "model_version": null
   },
-  "model_family_id": "google/gemini-3-flash",
+  "model_group_id": "google/gemini-3-flash",
   "model_route_id": "google__gemini-3-flash",
   "model_family_name": "Gemini 3 Flash Preview (Non-reasoning)",
   "raw_model_ids": [

tests/fixtures/models/openai__gpt-5-2-pro.json

@@ -28,7 +28,7 @@
     "model_route_id": "openai__gpt-5-2-pro",
     "model_version": null
   },
-  "model_family_id": "openai/gpt-5-2-pro",
+  "model_group_id": "openai/gpt-5-2-pro",
   "model_route_id": "openai__gpt-5-2-pro",
   "model_family_name": "GPT-5.2 Pro",
   "raw_model_ids": [

tests/pipeline-contract.test.ts

@@ -100,19 +100,19 @@ describe("Tier A — pipeline contracts (model files)", () => {
     expect(violations, formatViolations(violations)).toEqual([])
   })
 
-  it("model card has model_family_id matching pipelineSlugify(model_family_id) → model_route_id", () => {
+  it("model card has model_group_id matching pipelineSlugify(model_group_id) → model_route_id", () => {
     const violations: Violation[] = []
     for (const { id, data } of models) {
-      if (!data.model_family_id) {
-        violations.push({ fixture: id, path: "model_family_id", detail: "missing" })
+      if (!data.model_group_id) {
+        violations.push({ fixture: id, path: "model_group_id", detail: "missing" })
         continue
       }
-      const expected = data.model_family_id.replace(/\//g, "__")
+      const expected = data.model_group_id.replace(/\//g, "__")
       if (data.model_route_id !== expected) {
         violations.push({
           fixture: id,
           path: "model_route_id",
-          detail: `${data.model_route_id} !== ${expected} (derived from ${data.model_family_id})`,
+          detail: `${data.model_route_id} !== ${expected} (derived from ${data.model_group_id})`,
         })
       }
     }

tests/redirect-integrity.test.ts

@@ -0,0 +1,93 @@
+import { readFileSync } from "fs"
+import { fileURLToPath } from "url"
+
+import { DuckDBConnection } from "@duckdb/node-api"
+import { beforeAll, describe, expect, it } from "vitest"
+
+// Server-FREE data-contract checks for the model URL redirect map + resolver
+// fallback invariants, asserted against a Stage-J warehouse snapshot. Opt-in via
+// `SNAPSHOT_URL=<warehouse> pnpm test:integrity` — self-skips in the default
+// `pnpm test` run (mirrors the RUN_DRIFT pattern in upstream-drift.test.ts) so it
+// never blocks a dev or a deploy. The server-BOUND checks (page rendering, chart
+// content, redirect-preservation) live in the Playwright e2e suite + PREFLIGHT.md.
+//
+// Why this exists: "the page returns 200" != "correct". The redirect map and the
+// raw_model_ids fallback have invariants that, if violated, silently break model
+// pages (folded ids 404, an inverted map 301s working URLs to dead leaves). See
+// tests/PREFLIGHT.md for the full bug taxonomy.
+
+const SNAPSHOT = (process.env.SNAPSHOT_URL || "").replace(/\/+$/, "")
+const shouldRun = !!SNAPSHOT
+const MAP_PATH = fileURLToPath(new URL("../lib/model-url-redirects.ts", import.meta.url))
+
+const dec = (s: string) => { try { return decodeURIComponent(s) } catch { return s } }
+const listItems = (v: unknown): string[] =>
+  Array.isArray(v) ? v.map(String)
+    : v && typeof v === "object" && Array.isArray((v as { items?: unknown[] }).items)
+      ? (v as { items: unknown[] }).items.map(String)
+      : []
+
+describe.skipIf(!shouldRun)("redirect-map + fallback integrity (vs SNAPSHOT_URL)", () => {
+  let map: [string, string][] = []
+  let keys: Set<string>
+  const addressable = new Set<string>() // decoded route forms that resolve to a page
+  const rawOwners = new Map<string, Set<string>>() // raw spelling -> owning group(s)
+  let nullRaw = 0
+  let rowCount = 0
+
+  beforeAll(async () => {
+    map = [...readFileSync(MAP_PATH, "utf8").matchAll(/\["([^"]+)",\s*"([^"]+)"\]/g)].map((m) => [m[1], m[2]])
+    keys = new Set(map.map(([k]) => k))
+    const con = await DuckDBConnection.create()
+    await con.run("INSTALL httpfs; LOAD httpfs;")
+    const rows = (await con.runAndReadAll(
+      `SELECT model_id, route_id, model_route_id, model_group_id, model_key, raw_model_ids
+       FROM read_parquet('${SNAPSHOT}/models_view.parquet')`,
+    )).getRowObjects()
+    rowCount = rows.length
+    for (const r of rows) {
+      for (const v of [r.model_id, r.model_route_id, r.route_id, r.model_group_id, r.model_key]) {
+        if (v) addressable.add(dec(String(v)))
+      }
+      if (r.raw_model_ids == null) nullRaw++
+      for (const raw of listItems(r.raw_model_ids)) {
+        const set = rawOwners.get(String(raw)) ?? rawOwners.set(String(raw), new Set()).get(String(raw))!
+        set.add(String(r.model_id))
+      }
+    }
+  }, 120_000)
+
+  it("redirect map has no self-redirects", () => {
+    expect(map.filter(([k, v]) => k === v)).toEqual([])
+  })
+
+  it("redirect map has no chains/loops (no target is also a key)", () => {
+    expect(map.filter(([, v]) => keys.has(v)).map(([, v]) => v)).toEqual([])
+  })
+
+  it("every redirect target is an addressable route", () => {
+    const dead = [...new Set(map.map(([, v]) => v))].filter((t) => !addressable.has(dec(t)))
+    expect(dead, `dead targets: ${dead.slice(0, 8)}`).toEqual([])
+  })
+
+  it("direction is folded->group (no addressable id is a redirect KEY)", () => {
+    // An addressable key means the map redirects a WORKING page away — the
+    // inverted-map bug.
+    const inverted = map.filter(([k]) => addressable.has(dec(k))).map(([k]) => k)
+    expect(inverted, `addressable ids used as redirect keys: ${inverted.slice(0, 8)}`).toEqual([])
+  })
+
+  it("every redirect key is a known folded raw_model_id", () => {
+    const orphan = map.filter(([k]) => !addressable.has(dec(k)) && !rawOwners.has(dec(k))).map(([k]) => k)
+    expect(orphan, `orphan keys (neither addressable nor a raw id): ${orphan.slice(0, 8)}`).toEqual([])
+  })
+
+  it("no raw_model_id belongs to >1 group (LIMIT 1 fallback is unambiguous)", () => {
+    const multi = [...rawOwners].filter(([, s]) => s.size > 1).map(([k]) => k)
+    expect(multi, `raw ids in >1 group: ${multi.slice(0, 8)}`).toEqual([])
+  })
+
+  it("no models_view row has NULL raw_model_ids (the fallback scans this column)", () => {
+    expect(nullRaw, `${nullRaw}/${rowCount} rows have NULL raw_model_ids`).toBe(0)
+  })
+})

tests/transformations/identity-canonicalization.test.ts

@@ -1,251 +0,0 @@
-import { describe, expect, it } from "vitest"
-
-import { getCanonicalModelIdentity, getModelFamilyRouteId } from "../../lib/model-family"
-
-// Executable spec for the identity-canonicalization transformation.
-//
-// Each table here corresponds to a Group in the spec. Pipeline-side
-// implementation must produce identical outputs for every row. Adding a row
-// here is the way to extend the spec.
-//
-// When the pipeline ships matching values, run `scripts/verify-identity.mjs`
-// against the full live cache to confirm cross-corpus equivalence before
-// deleting the TS implementation.
-
-interface Case {
-  id: string
-  name?: string
-  developer?: string
-  expected: {
-    namespace?: string
-    familySlug?: string
-    familyId?: string
-    familyName?: string
-    variantKey?: string
-    variantLabel?: string
-    versionDate?: string | undefined
-    versionQualifier?: string | undefined
-    variantDisplayName?: string
-  }
-  why: string
-}
-
-// ---------------------------------------------------------------------------
-// Group A — Token case map
-// ---------------------------------------------------------------------------
-//
-// Each token in the map below must title-case to the canonical form. The full
-// case map is captured in the spec doc (Group A table). Here we exercise a
-// representative subset plus the "no map entry" baseline.
-
-describe("Group A — token case map", () => {
-  const cases: Case[] = [
-    { id: "anthropic/claude-opus-4.5", name: "Claude Opus 4.5", expected: { familyName: "Claude Opus 4.5" }, why: "claude → Claude, opus → Opus" },
-    { id: "openai/gpt-5", name: "GPT 5", expected: { familyName: "GPT 5" }, why: "gpt → GPT (uppercased)" },
-    { id: "google/gemini-2.5-pro", name: "Gemini 2.5 Pro", expected: { familyName: "Gemini 2.5 Pro" }, why: "gemini, pro" },
-    { id: "meta/llama-3", name: "Llama 3", expected: { familyName: "Llama 3" }, why: "llama" },
-    { id: "01-ai/yi-large", name: "Yi Large", expected: { familyName: "Yi Large" }, why: "yi (lowercase letter, but in map)" },
-    { id: "anthropic/claude-haiku", name: "Claude Haiku", expected: { familyName: "Claude Haiku" }, why: "haiku" },
-    { id: "anthropic/claude-instant", name: "Claude Instant", expected: { familyName: "Claude Instant" }, why: "instant" },
-    { id: "x/x-ai", name: "x AI", expected: { familyName: "X AI" }, why: "ai → AI (a deliberate capitalization)" },
-    { id: "test/turbo-mini", name: "Turbo Mini", expected: { familyName: "Turbo Mini" }, why: "turbo, mini" },
-    { id: "qwen/qwen-coder", name: "Qwen Coder", expected: { familyName: "Qwen Coder" }, why: "qwen, coder" },
-    { id: "test/foo-bar", name: "Foo Bar", expected: { familyName: "Foo Bar" }, why: "tokens not in map → naive title-case" },
-  ]
-  it.each(cases)("$why → familyName=$expected.familyName", ({ id, name, expected }) => {
-    const result = getCanonicalModelIdentity({ id, name: name ?? "" })
-    expect(result.familyName).toBe(expected.familyName)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group B — v/V version-token rule
-// ---------------------------------------------------------------------------
-//
-// Tokens matching /^v\d/i are lowercased. This rule is the source of the
-// 1,253-card pipeline disagreement documented in the spec; pipeline must
-// apply this rule before emitting model_family_name.
-
-describe("Group B — v/V version-token rule", () => {
-  const cases: Case[] = [
-    { id: "deepseek-ai/deepseek-v3", name: "Deepseek V3", expected: { familyName: "Deepseek v3" }, why: "trailing v3" },
-    { id: "deepseek-ai/deepseek-v3.1", name: "Deepseek V3.1", expected: { familyName: "Deepseek v3.1" }, why: "v3.1 with patch" },
-    { id: "mistralai/mistral-7b-instruct-v0.3", name: "Mistral 7B Instruct v0.3", expected: { familyName: "Mistral 7B Instruct v0.3" }, why: "v0.3 in middle, mistral & instruct in case map" },
-    { id: "mistralai/mixtral-8x22b-instruct-v0.1", name: "Mixtral 8x22b Instruct v0.1", expected: { familyName: "Mixtral 8x22b Instruct v0.1" }, why: "v0.1 with size token" },
-    { id: "amazon/nova-lite-v1.0", name: "Nova Lite v1.0", expected: { familyName: "Nova Lite v1.0" }, why: "v1.0" },
-    { id: "test/foo-bar", name: "Foo Bar", expected: { familyName: "Foo Bar" }, why: "no v rule applies" },
-  ]
-  it.each(cases)("$why → familyName=$expected.familyName", ({ id, name, expected }) => {
-    expect(getCanonicalModelIdentity({ id, name: name ?? "" }).familyName).toBe(expected.familyName)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group C — Date and qualifier extraction
-// ---------------------------------------------------------------------------
-
-describe("Group C — date and qualifier extraction", () => {
-  const cases: Case[] = [
-    {
-      id: "anthropic/claude-3-5-sonnet",
-      name: "Claude 3.5 Sonnet",
-      expected: { familySlug: "claude-3.5-sonnet", versionDate: undefined, versionQualifier: undefined, variantKey: "base", variantLabel: "Current" },
-      why: "no date in handle → base variant, Current label",
-    },
-    {
-      id: "anthropic/claude-3-5-sonnet-20240620",
-      name: "Claude 3.5 Sonnet 20240620",
-      expected: { familySlug: "claude-3.5-sonnet", versionDate: "2024-06-20", versionQualifier: undefined, variantKey: "20240620", variantLabel: "2024-06-20" },
-      why: "YYYYMMDD date present, no qualifier",
-    },
-    {
-      id: "anthropic/claude-3-5-sonnet-20240620-thinking",
-      name: "Claude 3.5 Sonnet 20240620 Thinking",
-      expected: {
-        familySlug: "claude-3.5-sonnet",
-        versionDate: "2024-06-20",
-        versionQualifier: "Thinking",
-        variantKey: "20240620-thinking",
-        variantLabel: "2024-06-20 · Thinking",
-      },
-      why: "date plus single-token qualifier",
-    },
-    {
-      id: "openai/gpt-5-2025-12-11-thinking-high",
-      name: "GPT 5 2025-12-11 thinking high",
-      expected: {
-        // The (?:19|20)\d{6} regex requires 8 contiguous digits; "2025-12-11" has dashes, so
-        // it doesn't match as a date. NormalizeHandle's digit-dash-digit collapse only fires when
-        // the right side is followed by a dash or end, so "2025-12-11" stays as is.
-        // → no date matched → base variant.
-        variantKey: "base",
-        versionDate: undefined,
-      },
-      why: "dashed date is NOT recognized — only YYYYMMDD form",
-    },
-  ]
-  it.each(cases)("$why", ({ id, name, expected }) => {
-    const result = getCanonicalModelIdentity({ id, name: name ?? "" })
-    if (expected.familySlug !== undefined) expect(result.familySlug).toBe(expected.familySlug)
-    if (expected.versionDate !== undefined || "versionDate" in expected) expect(result.versionDate).toBe(expected.versionDate)
-    if (expected.versionQualifier !== undefined || "versionQualifier" in expected) expect(result.versionQualifier).toBe(expected.versionQualifier)
-    if (expected.variantKey !== undefined) expect(result.variantKey).toBe(expected.variantKey)
-    if (expected.variantLabel !== undefined) expect(result.variantLabel).toBe(expected.variantLabel)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group D — Handle normalization
-// ---------------------------------------------------------------------------
-
-describe("Group D — handle normalization (familyId derivation)", () => {
-  const cases: Case[] = [
-    { id: "anthropic/Claude_Opus_4.5", name: "Claude Opus 4.5", expected: { familyId: "anthropic/claude-opus-4.5" }, why: "underscore → dash, lowercase" },
-    { id: "anthropic/claude opus 4.5", name: "Claude Opus 4.5", expected: { familyId: "anthropic/claude-opus-4.5" }, why: "space → dash" },
-    { id: "anthropic/--claude--opus--", name: "Claude Opus", expected: { familyId: "anthropic/claude-opus" }, why: "leading/trailing/repeated dashes collapsed" },
-    { id: "anthropic/claude-3-5-sonnet", name: "Claude 3.5 Sonnet", expected: { familyId: "anthropic/claude-3.5-sonnet" }, why: "digit-dash-digit-dash collapses to digit-dot-digit-dash" },
-    { id: "anthropic/claude-3-5", name: "Claude 3.5", expected: { familyId: "anthropic/claude-3.5" }, why: "digit-dash-digit at end collapses to digit-dot-digit" },
-    { id: "google/gemini-1.5-pro", name: "Gemini 1.5 Pro", expected: { familyId: "google/gemini-1.5-pro" }, why: "existing dot is preserved" },
-  ]
-  it.each(cases)("$why", ({ id, name, expected }) => {
-    expect(getCanonicalModelIdentity({ id, name: name ?? "" }).familyId).toBe(expected.familyId)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group E — Namespace and rawHandle extraction
-// ---------------------------------------------------------------------------
-
-describe("Group E — namespace + rawHandle extraction", () => {
-  it("id with slash → first segment is namespace, rest is handle", () => {
-    const r = getCanonicalModelIdentity({ id: "anthropic/claude-opus-4-5", name: "Claude Opus 4.5" })
-    expect(r.namespace).toBe("anthropic")
-    expect(r.rawHandle).toBe("claude-opus-4-5")
-  })
-  it("id without slash falls back to developer field for namespace", () => {
-    const r = getCanonicalModelIdentity({ id: "gpt-5", name: "GPT 5", developer: "OpenAI" })
-    expect(r.namespace).toBe("openai")
-  })
-  it("developer with spaces is slug-cased for namespace", () => {
-    const r = getCanonicalModelIdentity({ id: "claude-opus-4.5", name: "Claude Opus 4.5", developer: "Anthropic AI" })
-    expect(r.namespace).toBe("anthropic-ai")
-  })
-  it("namespace prefix in name is stripped when computing rawHandle from name", () => {
-    // Falls into getRawHandle's strippedName path when id has no '/' segment
-    const r = getCanonicalModelIdentity({ id: "", name: "anthropic/Claude Opus 4.5", developer: "Anthropic" })
-    expect(r.namespace).toBe("anthropic")
-    // Note: with id empty, idHandle is empty, so falls back to stripNamespace(name, namespace)
-    expect(r.rawHandle).toBe("Claude Opus 4.5")
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group F — familyId and model_route_id contract
-// ---------------------------------------------------------------------------
-
-describe("Group F — familyId and model_route_id", () => {
-  // KNOWN-RED (intentional): getModelFamilyRouteId now returns percent-encoded
-  // ids (e.g. `openai%2Fgpt-5`) — the correct v2 form, matching the warehouse
-  // `route_id`. These cases still assert the legacy `__` form.
-  // TODO: clean up route_id upstream (Stage J producer) to slash-form, then
-  // align getModelFamilyRouteId + these expectations.
-  const cases = [
-    { familyId: "anthropic/claude-opus-4.5", routeId: "anthropic__claude-opus-4.5" },
-    { familyId: "openai/gpt-5", routeId: "openai__gpt-5" },
-    { familyId: "01-ai/yi-large", routeId: "01-ai__yi-large" },
-    { familyId: "google/gemini-2.5-pro", routeId: "google__gemini-2.5-pro" },
-  ]
-  it.each(cases)("getModelFamilyRouteId($familyId) === $routeId", ({ familyId, routeId }) => {
-    expect(getModelFamilyRouteId(familyId)).toBe(routeId)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Integration cases — full-tuple round-trips for fixtures we ship
-// ---------------------------------------------------------------------------
-//
-// These mirror real model cards in tests/fixtures/model-cards/ (where the
-// pipeline ALSO emits the canonical fields). When pipeline matches across the
-// full corpus, this section is a sanity check that the spec stays in sync
-// with the fixtures.
-
-describe("integration — fixtures round-trip", () => {
-  const fixtures = [
-    {
-      id: "anthropic/claude-opus-4.5",
-      name: "Claude Opus 4.5",
-      expected: {
-        familyId: "anthropic/claude-opus-4.5",
-        familyName: "Claude Opus 4.5",
-        variantKey: "base",
-      },
-      why: "anthropic__claude-opus-4.5 fixture (dotted route_id quirk)",
-    },
-    {
-      id: "openai/gpt-5",
-      name: "GPT 5",
-      expected: {
-        familyId: "openai/gpt-5",
-        familyName: "GPT 5",
-        variantKey: "base",
-      },
-      why: "openai__gpt-5 fixture",
-    },
-    {
-      id: "01-ai/yi-34b",
-      name: "Yi 34B",
-      expected: {
-        familyId: "01-ai/yi-34b",
-        familyName: "Yi 34B",
-        variantKey: "base",
-      },
-      why: "01-ai__yi-34b fixture (dash-prefix slug)",
-    },
-  ]
-  it.each(fixtures)("$why", ({ id, name, expected }) => {
-    const result = getCanonicalModelIdentity({ id, name })
-    expect(result.familyId).toBe(expected.familyId)
-    expect(result.familyName).toBe(expected.familyName)
-    expect(result.variantKey).toBe(expected.variantKey)
-  })
-})

tests/transformations/model-url-redirects.test.ts

@@ -0,0 +1,219 @@
+import { describe, expect, it } from "vitest"
+
+import {
+  buildRedirectsFromModelsView,
+  routeId,
+  serializeRedirectModule,
+  type ModelsViewRow,
+} from "@/lib/model-url-redirects-build"
+import { resolveModelRedirect } from "@/lib/model-url-redirects"
+import { routeIdFromModelId, routeIdToPath } from "@/lib/utils"
+
+// model-resolution-rework — executable spec for the model-URL redirect layer.
+//
+// Covers the two CRITICAL contract points the playbook calls out:
+//   1. percent-encode / round-trip of model ids (RFC 3986, `/` -> %2F).
+//   2. query-param preservation across the 301 (asserted via the middleware's
+//      target-URL construction, replicated here to stay node-only / no Next
+//      runtime).
+
+// ---------------------------------------------------------------------------
+// Percent-encoding round-trip
+// ---------------------------------------------------------------------------
+
+describe("route id percent-encoding round-trips", () => {
+  const ids = [
+    "anthropic/claude-3.5-sonnet-20240620",
+    "meta/Llama-3.1-8B",
+    "01-ai/Yi-1.5-34B",
+    "EleutherAI/pythia-6.9b",
+    "ByteDance-Seed/Seed-OSS-36B-Instruct",
+    "openai/gpt-4o",
+    "mistralai/mistral-7b-instruct-v0.3",
+  ]
+
+  it.each(ids)("encode→decode is identity for %s", (id) => {
+    const encoded = routeIdFromModelId(id)
+    // No bare slashes survive encoding (single-segment route id form).
+    expect(encoded).not.toContain("/")
+    // Full round-trip back to the original id.
+    expect(decodeURIComponent(encoded)).toBe(id)
+  })
+
+  it("routeIdToPath turns the encoded form back into a slash path", () => {
+    const encoded = routeIdFromModelId("anthropic/claude-3.5-sonnet-20240620")
+    expect(encoded).toBe("anthropic%2Fclaude-3.5-sonnet-20240620")
+    expect(routeIdToPath(encoded)).toBe("anthropic/claude-3.5-sonnet-20240620")
+  })
+
+  it("routeIdFromModelId trims and tolerates empty/nullish", () => {
+    expect(routeIdFromModelId("  openai/gpt-4o  ")).toBe("openai%2Fgpt-4o")
+    expect(routeIdFromModelId(undefined)).toBe("")
+    expect(routeIdFromModelId(null)).toBe("")
+    expect(routeIdFromModelId("")).toBe("")
+  })
+
+  it("casing is preserved through encoding (HF ids are case-sensitive)", () => {
+    expect(routeIdFromModelId("meta/Llama-3.1-8B")).toBe("meta%2FLlama-3.1-8B")
+    expect(routeIdFromModelId("meta/llama-3.1-8b")).not.toBe(
+      routeIdFromModelId("meta/Llama-3.1-8B"),
+    )
+  })
+})
+
+// ---------------------------------------------------------------------------
+// buildRedirectsFromModelsView — map construction from the warehouse
+// ---------------------------------------------------------------------------
+
+describe("buildRedirectsFromModelsView", () => {
+  it("maps a folded raw spelling to its owning group route", () => {
+    const rows: ModelsViewRow[] = [
+      {
+        model_route_id: "mistralai%2Fmistral-medium",
+        model_id: "mistralai/mistral-medium",
+        route_id: "mistralai%2Fmistral-medium",
+        model_key: "mistralai/mistral-medium",
+        model_group_id: "mistralai/mistral-medium",
+        raw_model_ids: ["mistralai/mistral-medium-2505", "mistralai/Mistral-Medium"],
+      },
+    ]
+    const { redirects } = buildRedirectsFromModelsView(rows)
+    expect(redirects.get(routeId("mistralai/mistral-medium-2505"))).toBe("mistralai%2Fmistral-medium")
+    expect(redirects.get(routeId("mistralai/Mistral-Medium"))).toBe("mistralai%2Fmistral-medium")
+  })
+
+  it("never redirects an addressable id — no group->leaf hijack", () => {
+    const rows: ModelsViewRow[] = [
+      {
+        model_route_id: "mistralai%2Fmistral-medium",
+        model_id: "mistralai/mistral-medium",
+        // the group's own id appearing in its raw list must NOT become a redirect
+        raw_model_ids: ["mistralai/mistral-medium"],
+      },
+    ]
+    const { redirects } = buildRedirectsFromModelsView(rows)
+    expect(redirects.size).toBe(0)
+  })
+
+  it("every redirect target is an addressable model_route_id", () => {
+    const rows: ModelsViewRow[] = [
+      { model_route_id: "org%2Fgroup-a", model_id: "org/group-a", raw_model_ids: ["org/snap-1"] },
+      { model_route_id: "org%2Fgroup-b", model_id: "org/group-b", raw_model_ids: ["org/snap-2"] },
+    ]
+    const { redirects } = buildRedirectsFromModelsView(rows)
+    const addressable = new Set(rows.map((r) => r.model_route_id))
+    for (const target of redirects.values()) {
+      expect(addressable.has(target)).toBe(true)
+    }
+  })
+
+  it("excludes (does not arbitrarily pick) a spelling that fans out to multiple groups", () => {
+    const rows: ModelsViewRow[] = [
+      { model_route_id: "org%2Fgroup-a", model_id: "org/group-a", raw_model_ids: ["org/shared"] },
+      { model_route_id: "org%2Fgroup-b", model_id: "org/group-b", raw_model_ids: ["org/shared"] },
+    ]
+    const { redirects, ambiguous } = buildRedirectsFromModelsView(rows)
+    expect(redirects.has(routeId("org/shared"))).toBe(false)
+    expect(ambiguous.get(routeId("org/shared"))?.size).toBe(2)
+  })
+
+  it("is idempotent when the same spelling appears repeatedly", () => {
+    const rows: ModelsViewRow[] = [
+      { model_route_id: "org%2Fg", model_id: "org/g", raw_model_ids: ["org/snap", "org/snap"] },
+    ]
+    const { redirects } = buildRedirectsFromModelsView(rows)
+    expect(redirects.size).toBe(1)
+    expect(redirects.get(routeId("org/snap"))).toBe("org%2Fg")
+  })
+})
+
+// ---------------------------------------------------------------------------
+// serializeRedirectModule — generated source shape
+// ---------------------------------------------------------------------------
+
+describe("serializeRedirectModule", () => {
+  it("produces sorted, parseable entries derived from the warehouse", () => {
+    const map = new Map([
+      ["b%2Fx", "b%2Fy"],
+      ["a%2Fx", "a%2Fy"],
+    ])
+    const src = serializeRedirectModule(map, { source: "warehouse/test" })
+    expect(src).toContain("Derived from the warehouse")
+    // sorted: a before b
+    expect(src.indexOf('"a%2Fx"')).toBeLessThan(src.indexOf('"b%2Fx"'))
+    expect(src).toContain("export const MODEL_URL_REDIRECTS")
+    expect(src).toContain("export function resolveModelRedirect")
+  })
+})
+
+// ---------------------------------------------------------------------------
+// Redirect target construction — query-param preservation (CRITICAL)
+// ---------------------------------------------------------------------------
+//
+// Replicates the middleware target-building logic against a real URL so we
+// assert the 301 contract without booting the Next runtime.
+
+function buildRedirectTarget(incomingUrl: string): { url: URL; status: number } | null {
+  const url = new URL(incomingUrl)
+  const MODELS_PREFIX = "/models/"
+  if (!url.pathname.startsWith(MODELS_PREFIX)) return null
+  const rawTail = url.pathname.slice(MODELS_PREFIX.length)
+  if (!rawTail) return null
+  let decoded: string
+  try {
+    decoded = decodeURIComponent(rawTail)
+  } catch {
+    decoded = rawTail
+  }
+  const incomingRouteId = encodeURIComponent(decoded)
+  const newRouteId = resolveModelRedirect(incomingRouteId)
+  if (!newRouteId) return null
+  const target = new URL(url.toString())
+  target.pathname = `${MODELS_PREFIX}${newRouteId}`
+  target.search = url.search // preserve query params
+  return { url: target, status: 301 }
+}
+
+describe("redirect target construction (middleware contract)", () => {
+  // A known folded-spelling -> group redirect from the generated map: the dated
+  // snapshot folds into the moving group pointer.
+  const OLD = "anthropic/claude-3-haiku-20240307"
+  const NEW = "anthropic/claude-3-haiku"
+
+  it("the generated map contains the expected folded-spelling -> group redirect", () => {
+    expect(resolveModelRedirect(routeId(OLD))).toBe(routeId(NEW))
+  })
+
+  it("redirects an old encoded URL to the new one with 301", () => {
+    const res = buildRedirectTarget(`https://x.test/models/${routeId(OLD)}`)
+    expect(res).not.toBeNull()
+    expect(res!.status).toBe(301)
+    expect(res!.url.pathname).toBe(`/models/${routeId(NEW)}`)
+  })
+
+  it("PRESERVES query params across the redirect", () => {
+    const res = buildRedirectTarget(`https://x.test/models/${routeId(OLD)}?version=20240307&foo=bar`)
+    expect(res).not.toBeNull()
+    expect(res!.url.search).toBe("?version=20240307&foo=bar")
+    expect(res!.url.searchParams.get("version")).toBe("20240307")
+    expect(res!.url.searchParams.get("foo")).toBe("bar")
+  })
+
+  it("preserves an empty query string (no spurious ?)", () => {
+    const res = buildRedirectTarget(`https://x.test/models/${routeId(OLD)}`)
+    expect(res!.url.search).toBe("")
+  })
+
+  it("handles the decoded (slash) path form too — round-trips to the same redirect", () => {
+    // Next.js may hand us the decoded path; the lookup re-encodes it.
+    const res = buildRedirectTarget(`https://x.test/models/${OLD}?version=v1`)
+    expect(res).not.toBeNull()
+    expect(res!.url.pathname).toBe(`/models/${routeId(NEW)}`)
+    expect(res!.url.searchParams.get("version")).toBe("v1")
+  })
+
+  it("returns null (no redirect) for an unknown model route", () => {
+    const res = buildRedirectTarget("https://x.test/models/unknown%2Fmodel-xyz")
+    expect(res).toBeNull()
+  })
+})

tests/transformations/setup-alias-merging.test.ts

@@ -1,268 +0,0 @@
-import { describe, expect, it } from "vitest"
-
-import { getCanonicalModelIdentity } from "../../lib/model-family"
-
-// Executable spec for the setup-alias-merging transformation.
-//
-// These tests describe TS as it currently runs in production. Quirks are
-// preserved on purpose — the migration target is to move the computation
-// upstream without changing what users see, not to fix transformation
-// decisions. If a test below looks "wrong" to product sense, that's a
-// future product decision; fixing it is out of scope here.
-//
-// Pipeline-side implementation must produce identical outputs for every
-// row. Verify cross-corpus equivalence with `scripts/verify-setup-alias.mjs`
-// once pipeline ships.
-
-// ---------------------------------------------------------------------------
-// Group A — isSetupAliasQualifier truth table
-// ---------------------------------------------------------------------------
-//
-// Reproduces the function from lib/hf-data.ts (and its identical
-// twin in scripts/cache-hf-data.mjs). Pipeline must match exactly.
-
-function normalizeSetupAliasQualifier(value: string | null | undefined): string {
-  return value?.trim().toLowerCase().replace(/[_\s]+/g, "-") ?? ""
-}
-
-function isSetupAliasQualifier(value: string | null | undefined): boolean {
-  const normalized = normalizeSetupAliasQualifier(value)
-  return (
-    normalized === "prompt" ||
-    normalized === "fc" ||
-    normalized === "function-calling" ||
-    normalized.startsWith("thinking")
-  )
-}
-
-describe("Group A — isSetupAliasQualifier", () => {
-  const cases = [
-    { input: "prompt", expected: true, why: "exact: prompt" },
-    { input: "Prompt", expected: true, why: "case-insensitive" },
-    { input: "PROMPT", expected: true, why: "case-insensitive" },
-    { input: "fc", expected: true, why: "exact: fc" },
-    { input: "FC", expected: true, why: "case-insensitive" },
-    { input: "function-calling", expected: true, why: "exact" },
-    { input: "function calling", expected: true, why: "space → dash" },
-    { input: "function_calling", expected: true, why: "underscore → dash" },
-    { input: "thinking", expected: true, why: "exact thinking" },
-    { input: "thinking-1k", expected: true, why: "starts with thinking" },
-    { input: "thinking-medium", expected: true, why: "starts with thinking" },
-    { input: "thinking-none", expected: true, why: "starts with thinking" },
-    { input: "thinking_xhigh", expected: true, why: "underscore → dash, then prefix" },
-    { input: "Thinking 1K", expected: true, why: "case + space normalized → starts with thinking" },
-    { input: "high", expected: false, why: "non-alias inference qualifier" },
-    { input: "medium", expected: false, why: "non-alias" },
-    { input: "low", expected: false, why: "non-alias" },
-    { input: "minimal", expected: false, why: "non-alias" },
-    { input: "8k", expected: false, why: "context-length without thinking prefix" },
-    { input: "16k", expected: false, why: "context-length without thinking prefix" },
-    { input: "", expected: false, why: "empty" },
-    { input: null, expected: false, why: "null" },
-    { input: undefined, expected: false, why: "undefined" },
-    { input: "  prompt  ", expected: true, why: "leading/trailing whitespace trimmed" },
-    { input: "prompts", expected: false, why: "trailing s — not exact prompt and no thinking prefix" },
-    { input: "fcfc", expected: false, why: "doesn't match exact fc" },
-  ]
-  it.each(cases)("'$input' → $expected ($why)", ({ input, expected }) => {
-    expect(isSetupAliasQualifier(input)).toBe(expected)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group B — End-to-end variant normalization (TS-as-is)
-// ---------------------------------------------------------------------------
-//
-// Replicates lib/hf-data.ts verbatim, NO date-format fix applied.
-// Documents the dashed-date fall-through behaviour as the canonical spec.
-
-interface VariantInput {
-  variant_key: string
-  variant_label?: string
-}
-
-function normalizeOne(familyId: string, variant: VariantInput): { variant_key: string; variant_label: string } {
-  if (variant.variant_key === "base") {
-    return { variant_key: "default", variant_label: "Default" }
-  }
-  if (variant.variant_key === "default") {
-    return { variant_key: "default", variant_label: variant.variant_label ?? "Default" }
-  }
-
-  const synth = getCanonicalModelIdentity({
-    id: `${familyId}-${variant.variant_key}`,
-    name: `${familyId}-${variant.variant_key}`,
-  })
-
-  if (synth.versionDate && isSetupAliasQualifier(synth.versionQualifier)) {
-    return { variant_key: synth.versionDate, variant_label: synth.versionDate }
-  }
-  return { variant_key: synth.variantKey, variant_label: synth.variantLabel }
-}
-
-describe("Group B — End-to-end variant normalization", () => {
-  const familyId = "openai/gpt-5.2"
-  const cases = [
-    { variant_key: "default", expected: { variant_key: "default", variant_label: "Default" }, why: "default passes through" },
-    { variant_key: "base", expected: { variant_key: "default", variant_label: "Default" }, why: "base renamed to default" },
-    { variant_key: "20251101", expected: { variant_key: "20251101", variant_label: "2025-11-01" }, why: "YYYYMMDD date-only — preserved as raw token, ISO label" },
-    {
-      variant_key: "2025-11-01",
-      expected: { variant_key: "base", variant_label: "Current" },
-      why: "DASHED date-only falls through to base — TS quirk, preserved as canonical for this migration",
-    },
-    {
-      variant_key: "20240620-thinking",
-      expected: { variant_key: "2024-06-20", variant_label: "2024-06-20" },
-      why: "YYYYMMDD + thinking → merge to ISO date",
-    },
-    {
-      variant_key: "20240620-thinking-1k",
-      expected: { variant_key: "2024-06-20", variant_label: "2024-06-20" },
-      why: "YYYYMMDD + thinking-1k → merge (startsWith match aggregates all thinking budgets)",
-    },
-    {
-      variant_key: "20240620-thinking-medium",
-      expected: { variant_key: "2024-06-20", variant_label: "2024-06-20" },
-      why: "all thinking-N variants for this YYYYMMDD date collapse together",
-    },
-    { variant_key: "20240620-fc", expected: { variant_key: "2024-06-20", variant_label: "2024-06-20" }, why: "YYYYMMDD + fc → merge" },
-    { variant_key: "20240620-prompt", expected: { variant_key: "2024-06-20", variant_label: "2024-06-20" }, why: "YYYYMMDD + prompt → merge" },
-    {
-      variant_key: "20240620-high",
-      expected: { variant_key: "20240620-high", variant_label: "2024-06-20 · High" },
-      why: "non-alias qualifier preserved with date",
-    },
-    {
-      variant_key: "2025-12-11-thinking-medium",
-      expected: { variant_key: "base", variant_label: "Current" },
-      why: "DASHED date — regex doesn't match, falls through to base (TS quirk)",
-    },
-    {
-      variant_key: "2025-12-11-thinking-1k",
-      expected: { variant_key: "base", variant_label: "Current" },
-      why: "DASHED date with thinking-1k — same fall-through",
-    },
-    {
-      variant_key: "2025-12-11-fc",
-      expected: { variant_key: "base", variant_label: "Current" },
-      why: "DASHED date + fc — fall-through",
-    },
-    {
-      variant_key: "2025-12-11-high",
-      expected: { variant_key: "base", variant_label: "Current" },
-      why: "DASHED date + non-alias qualifier — fall-through",
-    },
-    { variant_key: "gpt-foo-bar", expected: { variant_key: "base", variant_label: "Current" }, why: "no date detected" },
-  ]
-  it.each(cases)("'$variant_key' → '$expected.variant_key' ($why)", ({ variant_key, expected }) => {
-    const result = normalizeOne(familyId, { variant_key })
-    expect(result.variant_key).toBe(expected.variant_key)
-    expect(result.variant_label).toBe(expected.variant_label)
-  })
-})
-
-// ---------------------------------------------------------------------------
-// Group C — Multi-variant deduplication after normalization
-// ---------------------------------------------------------------------------
-//
-// Documents the user-visible aggregation effect: cards with multiple
-// dashed-date variants all collapse into a single "base" entry. This is
-// TS as-is. If the team later decides users would benefit from
-// disaggregation, that's a separate product call (see the spec doc).
-
-function normalizeVariants(
-  familyId: string,
-  variants: Array<VariantInput & { evaluation_count?: number; raw_model_ids?: string[]; last_updated?: string }>
-) {
-  const byKey = new Map<
-    string,
-    { variant_key: string; variant_label: string; evaluation_count: number; raw_model_ids: string[]; last_updated?: string }
-  >()
-  for (const v of variants) {
-    const norm = normalizeOne(familyId, v)
-    const existing = byKey.get(norm.variant_key)
-    if (existing) {
-      existing.evaluation_count += v.evaluation_count ?? 0
-      existing.raw_model_ids = Array.from(new Set([...existing.raw_model_ids, ...(v.raw_model_ids ?? [])])).sort()
-      if (v.last_updated && (!existing.last_updated || new Date(v.last_updated) > new Date(existing.last_updated))) {
-        existing.last_updated = v.last_updated
-      }
-    } else {
-      byKey.set(norm.variant_key, {
-        variant_key: norm.variant_key,
-        variant_label: norm.variant_label,
-        evaluation_count: v.evaluation_count ?? 0,
-        raw_model_ids: [...(v.raw_model_ids ?? [])].sort(),
-        last_updated: v.last_updated,
-      })
-    }
-  }
-  return [...byKey.values()]
-}
-
-describe("Group C — Multi-variant deduplication (TS-as-is)", () => {
-  it("openai/gpt-5.2: 7 dashed-date variants collapse into default + base", () => {
-    const result = normalizeVariants("openai/gpt-5.2", [
-      { variant_key: "default", evaluation_count: 1, raw_model_ids: ["openai/gpt-5.2"] },
-      {
-        variant_key: "2025-12-11",
-        evaluation_count: 3,
-        raw_model_ids: ["openai/gpt-5.2-2025-12-11", "openai/gpt-5-2-2025-12-11-fc", "openai/gpt-5-2-2025-12-11-prompt"],
-      },
-      { variant_key: "2025-12-11-thinking-medium", evaluation_count: 1, raw_model_ids: ["openai/gpt-5-2-2025-12-11-thinking-medium"] },
-      { variant_key: "2025-12-11-thinking-low", evaluation_count: 1, raw_model_ids: ["openai/gpt-5-2-2025-12-11-thinking-low"] },
-      { variant_key: "2025-12-11-thinking-high", evaluation_count: 1, raw_model_ids: ["openai/gpt-5-2-2025-12-11-thinking-high"] },
-      { variant_key: "2025-12-11-thinking-none", evaluation_count: 1, raw_model_ids: ["openai/gpt-5-2-2025-12-11-thinking-none"] },
-      { variant_key: "2025-12-11-thinking-xhigh", evaluation_count: 1, raw_model_ids: ["openai/gpt-5-2-2025-12-11-thinking-xhigh"] },
-    ])
-    // All 6 dashed-date variants collapse into a single "base" entry. This
-    // is TS-as-is behaviour and the canonical spec for this migration.
-    expect(result.map((v) => v.variant_key).sort()).toEqual(["base", "default"])
-    const base = result.find((v) => v.variant_key === "base")!
-    expect(base.evaluation_count).toBe(8)
-    expect(base.raw_model_ids.length).toBe(8)
-  })
-
-  it("anthropic/claude-haiku-4.5: YYYYMMDD-thinking-Nk variants merge into ISO date (startsWith match fires)", () => {
-    const result = normalizeVariants("anthropic/claude-haiku-4.5", [
-      { variant_key: "default", evaluation_count: 1, raw_model_ids: ["anthropic/claude-haiku-4.5"] },
-      { variant_key: "20251001", evaluation_count: 2, raw_model_ids: ["anthropic/claude-haiku-4-5-20251001", "anthropic/claude-haiku-4-5-20251001-fc"] },
-      { variant_key: "20251001-thinking-1k", evaluation_count: 1, raw_model_ids: ["anthropic/claude-haiku-4-5-20251001-thinking-1k"] },
-      { variant_key: "20251001-thinking-8k", evaluation_count: 1, raw_model_ids: ["anthropic/claude-haiku-4-5-20251001-thinking-8k"] },
-      { variant_key: "20251001-thinking-16k", evaluation_count: 1, raw_model_ids: ["anthropic/claude-haiku-4-5-20251001-thinking-16k"] },
-      { variant_key: "20251001-thinking-32k", evaluation_count: 1, raw_model_ids: ["anthropic/claude-haiku-4-5-20251001-thinking-32k"] },
-    ])
-    // YYYYMMDD-thinking-Nk variants merge into "2024-10-01" (ISO) via the
-    // startsWith("thinking") match. The base "20251001" stays as YYYYMMDD
-    // because it has no qualifier. So they DON'T merge with each other —
-    // different normalized keys ("20251001" vs "2025-10-01"). TS quirk.
-    const keys = result.map((v) => v.variant_key).sort()
-    expect(keys).toContain("default")
-    expect(keys).toContain("20251001")
-    expect(keys).toContain("2025-10-01")
-    expect(keys.length).toBe(3)
-    const merged = result.find((v) => v.variant_key === "2025-10-01")!
-    expect(merged.evaluation_count).toBe(4)
-    expect(merged.raw_model_ids.length).toBe(4)
-  })
-
-  it("non-alias qualifiers with YYYYMMDD dates preserved as separate variants", () => {
-    const result = normalizeVariants("openai/gpt-5", [
-      { variant_key: "default", evaluation_count: 1, raw_model_ids: [] },
-      { variant_key: "20250807", evaluation_count: 1, raw_model_ids: [] },
-      { variant_key: "20250807-high", evaluation_count: 1, raw_model_ids: [] },
-      { variant_key: "20250807-low", evaluation_count: 1, raw_model_ids: [] },
-      { variant_key: "20250807-medium", evaluation_count: 1, raw_model_ids: [] },
-      { variant_key: "20250807-minimal", evaluation_count: 1, raw_model_ids: [] },
-    ])
-    expect(result.map((v) => v.variant_key).sort()).toEqual([
-      "20250807",
-      "20250807-high",
-      "20250807-low",
-      "20250807-medium",
-      "20250807-minimal",
-      "default",
-    ])
-  })
-})

tests/upstream-drift.test.ts

@@ -95,13 +95,13 @@ describe.skipIf(!shouldRun)(`Tier A drift — live cache contracts (${modelFiles
     expect(violations, `${violations} eval-details without category. Examples: ${examples.join(", ")}`).toBe(0)
   })
 
-  it("every model card has model_route_id === pipelineSlugify(model_family_id)", () => {
+  it("every model card has model_route_id === pipelineSlugify(model_group_id)", () => {
     let violations = 0
     const examples: string[] = []
     const cardsPath = path.resolve(import.meta.dirname, "..", ".cache", "hf-data", "model-cards.json")
     const cards = JSON.parse(fs.readFileSync(cardsPath, "utf8"))
     for (const card of cards) {
-      const expected = (card.model_family_id || "").replace(/\//g, "__")
+      const expected = (card.model_group_id || "").replace(/\//g, "__")
       if (card.model_route_id !== expected) {
         violations += 1
         if (examples.length < 5) examples.push(`${card.model_route_id} (expected ${expected})`)

tests/view-data.test.ts

@@ -27,7 +27,7 @@ async function writeSyntheticStageJSnapshot(snapshotDir: string) {
         'openai/gpt-5' AS id,
         'openai%2Fgpt-5' AS route_id,
         'openai%2Fgpt-5' AS model_route_id,
-        'openai/gpt-5' AS model_family_id,
+        'openai/gpt-5' AS model_group_id,
         'GPT 5' AS model_name,
         'GPT 5' AS canonical_model_name,
         'GPT 5' AS model_family_name,

vitest.config.ts

@@ -15,6 +15,8 @@ export default defineConfig({
     // self-skips otherwise. We DON'T exclude the path here because vitest's
     // exclude wins over an explicit path arg, which would silently make
     // `pnpm test:drift` find zero tests.
-    exclude: ["**/node_modules/**", "**/dist/**"],
+    // tests/e2e/ is the Playwright (@playwright/test) suite — a different runner.
+    // Excluded so `pnpm test` (vitest) doesn't try to collect its *.spec.ts.
+    exclude: ["**/node_modules/**", "**/dist/**", "**/tests/e2e/**"],
   },
 })