Route peer-ranks fetch through SNAPSHOT_URL sidecar

The model-detail benchmark grid was hardcoded to fetch
peer-ranks.json from the dataset *root* on `main`. The rest of
the app reads from the pinned `SNAPSHOT_URL=warehouse/<id>/`
snapshot, so peer ranks could drift from the views they were
paired with — different eval IDs, totals, or models than the
comparison-index showed.

The producer now emits peer-ranks.json as a Stage J sidecar
(eval_cards_backend_pipeline ffbfe71). Add `fetchPeerRanks()` to
`lib/sidecars.ts` (unwraps the new `{generated_at, ranks}`
envelope; gracefully returns `{}` for older snapshots that 404),
expose it through `lib/hf-data.ts` + `/api/peer-ranks` +
`dashboard-data-client.ts`, and switch `loadPeerRanks()` in
benchmark-detail.tsx off the hardcoded URL.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

app/api/peer-ranks/route.ts

@@ -0,0 +1,22 @@
+import { NextResponse } from "next/server"
+
+import { fetchPeerRanks } from "@/lib/hf-data"
+
+export async function GET() {
+  try {
+    const ranks = await fetchPeerRanks()
+    return NextResponse.json(ranks, {
+      headers: {
+        // Per-snapshot file, recomputed by the producer per pipeline run.
+        // Hot-cache aggressively in the browser; HF's own CDN handles
+        // dataset-side caching for us.
+        "Cache-Control": "public, max-age=600, stale-while-revalidate=3600",
+      },
+    })
+  } catch (err) {
+    return NextResponse.json(
+      { error: err instanceof Error ? err.message : "failed to load peer ranks" },
+      { status: 500 }
+    )
+  }
+}

components/benchmark-detail.tsx

@@ -47,8 +47,10 @@ import type {
   ComparisonMetricEntry,
   ComparisonScoreEntry,
   EvalHierarchy,
+  PeerRanksMap,
   SubmissionAxis,
 } from "@/lib/backend-artifacts"
+import { fetchPeerRanks } from "@/lib/dashboard-data-client"
 import {
   buildHierarchyEvalIndex,
   type HierarchyEvalLocation,
@@ -1371,18 +1373,17 @@ function getGroupPeerRank(
   return best ?? (group.bestRankPosition != null ? { position: group.bestRankPosition, total: group.bestRankTotal ?? 0 } : null)
 }
 
-type PeerRanksMap = Record<string, Record<string, { position: number; total: number }>>
-
+// peer-ranks.json now ships as a sidecar inside the pinned `SNAPSHOT_URL`
+// snapshot (Stage J emits it alongside hierarchy.json / comparison-index.json
+// — see eval_cards_backend_pipeline commit ffbfe71). Routing through the
+// same `/api/peer-ranks` endpoint as the other sidecars keeps peer ranks
+// pinned to the snapshot the rest of the page is reading from, instead of
+// drifting to the unversioned `main`-branch copy at the dataset root.
 let peerRanksPromise: Promise<PeerRanksMap> | null = null
 
-const DATASET_PEER_RANKS_URL =
-  "https://huggingface.co/datasets/evaleval/card_backend/resolve/main/peer-ranks.json"
-
 function loadPeerRanks(): Promise<PeerRanksMap> {
   if (!peerRanksPromise) {
-    peerRanksPromise = fetch(DATASET_PEER_RANKS_URL)
-      .then((r) => (r.ok ? r.json() : {}))
-      .catch(() => ({}))
+    peerRanksPromise = fetchPeerRanks().catch(() => ({} as PeerRanksMap))
   }
   return peerRanksPromise
 }

lib/backend-artifacts.ts

@@ -467,3 +467,23 @@ export interface ComparisonIndex {
   evals: Record<string, ComparisonEvalEntry>
   by_model: Record<string, Record<string, Record<string, ComparisonByModelEntry>>>
 }
+
+// ---------------------------------------------------------------------------
+// peer-ranks.json — primary-metric peer rank per (eval, model)
+// ---------------------------------------------------------------------------
+
+/** Bare map shape consumed by the model-detail benchmark grid. */
+export type PeerRanksMap = Record<
+  string,
+  Record<string, { position: number; total: number }>
+>
+
+/** Wrapped sidecar payload emitted by the v2 producer. Older (unversioned)
+ *  publishings of peer-ranks.json at the dataset root were a bare map; the
+ *  v2 snapshot wraps it with the same `{generated_at, config_version, ...}`
+ *  envelope as the other sidecars. */
+export interface PeerRanksSidecar {
+  generated_at: string
+  config_version: number
+  ranks: PeerRanksMap
+}

lib/dashboard-data-client.ts

@@ -1,4 +1,10 @@
-import type { BackendManifestStatus, ComparisonIndex, CorpusAggregates, EvalHierarchy } from "@/lib/backend-artifacts"
+import type {
+  BackendManifestStatus,
+  ComparisonIndex,
+  CorpusAggregates,
+  EvalHierarchy,
+  PeerRanksMap,
+} from "@/lib/backend-artifacts"
 import { decorateHierarchyDerivedTags } from "@/lib/benchmark-tags"
 import type { BenchmarkEvaluationCardData } from "@/components/benchmark-evaluation-card"
 import type { HFEvalDetail } from "@/lib/hf-data"
@@ -104,3 +110,7 @@ export function fetchComparisonIndex() {
 export function fetchCorpusAggregates() {
   return fetchJson<CorpusAggregates>("/api/corpus-aggregates")
 }
+
+export function fetchPeerRanks() {
+  return fetchJson<PeerRanksMap>("/api/peer-ranks")
+}

lib/hf-data.ts

@@ -16,6 +16,7 @@ import type {
   HierarchyMetric,
   HierarchySlice,
   HierarchyTags,
+  PeerRanksMap,
   RowAnnotations,
   SignalSummaries,
 } from "@/lib/backend-artifacts"
@@ -971,6 +972,19 @@ export async function fetchComparisonIndex(): Promise<ComparisonIndex> {
   return fetchHFJson<ComparisonIndex>("comparison-index.json")
 }
 
+/**
+ * Per-(eval, model) primary-metric peer ranks. v2 reads the wrapped
+ * sidecar from the pinned snapshot; legacy reads the bare-map file
+ * historically published unversioned at the dataset root.
+ */
+export async function fetchPeerRanks(): Promise<PeerRanksMap> {
+  if (useViewLayerBackend()) {
+    return (await fetchSnapshotSidecars()).fetchPeerRanks()
+  }
+
+  return (await fetchHFJsonSafe<PeerRanksMap>("peer-ranks.json")) ?? {}
+}
+
 export async function fetchCorpusAggregates(): Promise<CorpusAggregates | null> {
   if (useViewLayerBackend()) {
     return (await fetchSnapshotSidecars()).fetchHeadline()

lib/sidecars.ts

@@ -5,6 +5,8 @@ import type {
   ComparisonIndex,
   CorpusAggregates,
   EvalHierarchy,
+  PeerRanksMap,
+  PeerRanksSidecar,
 } from "@/lib/backend-artifacts"
 
 let cache: {
@@ -12,6 +14,7 @@ let cache: {
   headline?: Promise<CorpusAggregates>
   hierarchy?: Promise<EvalHierarchy>
   comparisonIndex?: Promise<ComparisonIndex>
+  peerRanks?: Promise<PeerRanksMap>
 } = {}
 
 function getSnapshotUrl() {
@@ -65,6 +68,30 @@ export function fetchComparisonIndex(): Promise<ComparisonIndex> {
   ))
 }
 
+/**
+ * Per-(eval, model) primary-metric peer ranks from
+ * `warehouse/<snapshot>/peer-ranks.json`. Resolves to the bare
+ * `eval_summary_id → model_route_id → {position, total}` map the
+ * model-detail benchmark grid expects, so callers don't have to reach
+ * into `.ranks` themselves.
+ *
+ * Returns an empty map if the snapshot doesn't carry the file yet —
+ * the producer started emitting it as a Stage J sidecar in May 2026,
+ * so older pinned snapshots may 404. Logs a warning in that case rather
+ * than throwing so the rest of the page still renders.
+ */
+export function fetchPeerRanks(): Promise<PeerRanksMap> {
+  return (cache.peerRanks ??= fetchJson<PeerRanksSidecar>("peer-ranks.json")
+    .then((payload) => payload?.ranks ?? {})
+    .catch((err) => {
+      console.warn(
+        `[sidecars] peer-ranks.json not available on snapshot; ` +
+          `falling back to empty map. ${err instanceof Error ? err.message : String(err)}`,
+      )
+      return {} as PeerRanksMap
+    }))
+}
+
 /**
  * Fail fast on contract regressions. Comparison-index rows must carry
  * `family_id`; the model-page graph view collapses without it. See