Consolidate hierarchy terminology + handle v2 hierarchy shape

Two threads in one change.

(1) v2 backend snapshot ships families[].composites[].benchmarks[]
instead of families[].leaves[]. The family-table only walked .leaves,
so every aggregator family rendered as a single non-expandable row
and clicking it navigated to one arbitrary child (e.g. clicking BFCL
landed on bfcl-live).

(2) The codebase mixed "Suite/Subtask" labels with "Composite/Slice"
data fields. Consolidate to one term per level so the UI and the
schema agree.

Hierarchy: 5 levels. Family -> Composite -> Benchmark -> Slice ->
Metric. "Suite" and "Subtask" are gone everywhere a human reads them.
Backend-shipped JSON keys (subtask_key, subtask_name,
metric.scope === "subtask", composite_benchmark_*) are kept since
they are wire contracts.

Family-table fix
- collectLeafEntries now treats nested benchmarks as the primary
source: walks family.standalone_benchmarks, family.benchmarks, and
family.composites[].benchmarks[] before falling back to the legacy
family.leaves[]. Each nested benchmark's summary_eval_ids[0] becomes
the navigation target, with `${family_key}_${benchmark_key}` as a
fallback when summary_eval_ids is missing.
- Verified end-to-end against the v2 snapshot
(j-chim/temp_evalcard_backend/warehouse/latest): bfcl now renders
"4 benchmarks" with the chevron, and clicking expands to BFCL Live /
BFCL Multi-turn / BFCL Non-live / BFCL Web Search.

Domain extraction in /evals
- The familyDomains and domainCounts memos walk the same nested
benchmarks before falling back to leaves, so the "Filter by
domain" pill list populates correctly under v2.

Hierarchy view
- Home page hierarchy strip collapses Split + Subtask into a single
"Slice" tier (5 levels instead of 6). Lede, kicker, OG image,
layout metadata description all updated to "five-level" to match.

Terminology rename (UI labels + local code identifiers)
- "Suite" / "Suites" -> "Composite" / "Composites"
- "Subtask" / "subtask" -> "Slice" / "slice"
- Renamed local identifiers in benchmark-detail.tsx and
eval-detail.tsx: SuiteGroup -> CompositeGroup,
getSuiteBadgeMeta -> getCompositeBadgeMeta,
normalizeSuiteKey -> normalizeCompositeKey,
groupBySuite -> groupByComposite,
getSuiteKey/Name/DisplayName -> getComposite*,
suiteGroups -> compositeGroups,
categorySuiteSections -> categoryCompositeSections,
suiteBenchmarkScoreRanges -> compositeBenchmarkScoreRanges,
activeSubtaskTab -> activeSliceTab,
subtaskMetricCounts -> sliceMetricCounts,
singleMetricSubtaskTabs -> singleMetricSliceTabs,
hasSubtaskTabs -> hasSliceTabs,
showSubtaskTopline -> showSliceTopline,
subtaskLabel/Count/DrivenBenchmarkCount -> slice* equivalents,
BenchmarkVariant discriminant strings "subtask" / "setup+subtask"
-> "slice" / "setup+slice",
subtasksOpen -> slicesOpen.
- Reworded prose in app/page.tsx, app/about/page.tsx,
app/evals/page.tsx, app/evals/[id]/page.tsx, app/layout.tsx,
app/opengraph-image.tsx, components/eval-card.tsx,
components/eval-detail.tsx, components/benchmark-evaluation-card.tsx,
components/policy-overview.tsx, components/model-compare-dialog.tsx.

Backend wire fields kept untouched (would require coordinated backend
change): metric.scope === "subtask", metric.subtask_key,
metric.subtask_name, summary.subtasks, slice.subtask_*,
composite_benchmark_name, composite_benchmark_key. A spec for the v2
maintainer covering the remaining family-grouping work
(gaia / gpqa / caparena variants) was sent separately.

app/about/page.tsx

@@ -65,8 +65,8 @@ export default function AboutPage() {
                 "Risk-domain annotations on benchmarks (capability, robustness, safety, agentic risk, fairness) so policy readers can locate which deployment-relevant property a number speaks to.",
               ],
               [
-                "Six-level hierarchy",
-                "Family → Suite → Single benchmark → Split → Subtask → Metric. Every score resolves to an explicit path, so aggregate claims drill down to the evidence supporting them.",
+                "Five-level hierarchy",
+                "Family → Composite → Single benchmark → Slice → Metric. Every score resolves to an explicit path, so aggregate claims drill down to the evidence supporting them.",
               ],
             ].map(([h, p]) => (
               <div
@@ -146,7 +146,7 @@ export default function AboutPage() {
               ],
               [
                 "Comparability",
-                "Can two scores under the same benchmark be put side-by-side? Tracks split, subtask, metric variant, and unit harmonisation; flags rows that cannot be ranked together.",
+                "Can two scores under the same benchmark be put side-by-side? Tracks slice, metric variant, and unit harmonisation; flags rows that cannot be ranked together.",
               ],
             ].map(([h, p], i) => (
               <li
@@ -183,7 +183,7 @@ export default function AboutPage() {
             {[
               [
                 "Canonicalisation",
-                "Heterogeneous score reports — papers, model cards, leaderboards, blog posts — are normalised to (model, benchmark, split, subtask, metric, value, source) tuples. Model name aliases and benchmark version aliases are resolved against a curated mapping.",
+                "Heterogeneous score reports — papers, model cards, leaderboards, blog posts — are normalised to (model, benchmark, slice, metric, value, source) tuples. Model name aliases and benchmark version aliases are resolved against a curated mapping.",
               ],
               [
                 "Source attribution",

app/evals/[id]/page.tsx

@@ -135,7 +135,7 @@ export default function EvalDetailPage() {
 }
 
 // ---------------------------------------------------------------------------
-// Composite (suite) view — paper §3.2 "composite reporting unit"
+// Composite view — paper §3.2 "composite reporting unit"
 // Surfaces sub-benchmarks as a hairline grid and a per-model × per-metric
 // matrix table. Both modes (research / policy) share the same chrome; the
 // policy-note panel changes per benchmark, surfaced from the sub-summary card.
@@ -169,8 +169,8 @@ function CompositeEvalView({
   const audience = card?.purpose_and_intended_users?.audience
   const audienceText = Array.isArray(audience) ? audience.join("; ") : audience
   const lede = isPolicy
-    ? overview || goal || `Suite aggregating ${subBenchmarkCount} component benchmarks across ${summary.models_count.toLocaleString()} models.`
-    : goal || overview || `Suite aggregating ${subBenchmarkCount} component benchmarks across ${summary.models_count.toLocaleString()} models.`
+    ? overview || goal || `Composite aggregating ${subBenchmarkCount} component benchmarks across ${summary.models_count.toLocaleString()} models.`
+    : goal || overview || `Composite aggregating ${subBenchmarkCount} component benchmarks across ${summary.models_count.toLocaleString()} models.`
 
   return (
     <div className="space-y-10">
@@ -281,7 +281,7 @@ function CompositeEvalView({
           style={{ color: "var(--fg-muted)", maxWidth: 720 }}
         >
           {tab === "metrics"
-            ? "Each card is one component benchmark inside this suite. Click a card to inspect its leaderboard, sub-tasks and benchmark card."
+            ? "Each card is one component benchmark inside this composite. Click a card to inspect its leaderboard, slices and benchmark card."
             : "Per-model scores across every component metric. Each column is a separately reported measure — distinct measures stay separate instead of collapsing into one number."}
         </p>
 

app/evals/page.tsx

@@ -98,11 +98,31 @@ export default function EvalsPage() {
     }
     for (const fam of families) {
       const seen = new Set<string>()
+      // Family-level fallback first (cards keyed by family slug).
       for (const d of lookupDomains(fam.key)) seen.add(d.trim().toLowerCase())
+
+      // v2 primary path: walk every nested benchmark across composites,
+      // standalone benchmarks, and any family-level benchmarks array.
+      const nestedBenchmarks = [
+        ...(fam.standalone_benchmarks ?? []),
+        ...(fam.benchmarks ?? []),
+        ...(fam.composites ?? []).flatMap((c) => c.benchmarks ?? []),
+      ]
+      for (const benchmark of nestedBenchmarks) {
+        for (const d of benchmark.tags?.domains ?? []) seen.add(d.trim().toLowerCase())
+        for (const d of lookupDomains(benchmark.key)) seen.add(d.trim().toLowerCase())
+        for (const id of benchmark.summary_eval_ids ?? []) {
+          for (const d of lookupDomains(id)) seen.add(d.trim().toLowerCase())
+        }
+      }
+
+      // Legacy fallback: per-leaf tags + cards keyed by leaf slug.
       for (const leaf of fam.leaves ?? []) {
         for (const d of leaf.tags?.domains ?? []) seen.add(d.trim().toLowerCase())
         for (const d of lookupDomains(leaf.key)) seen.add(d.trim().toLowerCase())
       }
+
+      // Family-level eval_summary_ids cover both shapes.
       for (const id of fam.eval_summary_ids ?? []) {
         for (const d of lookupDomains(id)) seen.add(d.trim().toLowerCase())
       }
@@ -126,6 +146,14 @@ export default function EvalsPage() {
       for (const d of card?.benchmark_details?.domains ?? []) recordLabel(d)
     }
     for (const fam of families) {
+      const nestedBenchmarks = [
+        ...(fam.standalone_benchmarks ?? []),
+        ...(fam.benchmarks ?? []),
+        ...(fam.composites ?? []).flatMap((c) => c.benchmarks ?? []),
+      ]
+      for (const benchmark of nestedBenchmarks) {
+        for (const d of benchmark.tags?.domains ?? []) recordLabel(d)
+      }
       for (const leaf of fam.leaves ?? []) {
         for (const d of leaf.tags?.domains ?? []) recordLabel(d)
       }
@@ -244,7 +272,7 @@ export default function EvalsPage() {
               </span>
             </div>
             <div className="ec-page-meta-item">
-              <span className="ec-page-meta-item-l">Suites</span>
+              <span className="ec-page-meta-item-l">Composites</span>
               <span className="ec-page-meta-item-v">
                 {stats.composite_count.toLocaleString()}
               </span>

app/layout.tsx

@@ -9,7 +9,7 @@ const SITE_URL = "https://evalcards.evalevalai.com"
 const SITE_NAME = "Eval Cards"
 const SITE_TITLE = "Eval Cards — a reporting layer for AI evaluations"
 const SITE_DESCRIPTION =
-  "A public registry of reported model–benchmark results, organised under a six-level rollout hierarchy and four interpretive signals: reproducibility, completeness, provenance, and comparability."
+  "A public registry of reported model–benchmark results, organised under a five-level rollout hierarchy and four interpretive signals: reproducibility, completeness, provenance, and comparability."
 
 export const metadata: Metadata = {
   metadataBase: new URL(SITE_URL),

app/opengraph-image.tsx

@@ -110,7 +110,7 @@ export default function OpenGraphImage() {
               maxWidth: "920px",
             }}
           >
-            A registry of reported model–benchmark results, organised under a six-level
+            A registry of reported model–benchmark results, organised under a five-level
             rollout hierarchy and four interpretive signals.
           </div>
         </div>

app/page.tsx

@@ -106,7 +106,7 @@ export default async function HomePage() {
             </h1>
             <p className="home-hero-lede">
               <strong>Eval Cards</strong> is a registry of reported model–benchmark results,
-              organised under a six-level rollout hierarchy and four interpretive signals
+              organised under a five-level rollout hierarchy and four interpretive signals
               computed over the joined record.
             </p>
             <div className="flex flex-wrap items-center gap-3">
@@ -164,9 +164,9 @@ export default async function HomePage() {
           </div>
         </section>
 
-        {/* SIX-LEVEL HIERARCHY STRIP -------------------------------------- */}
+        {/* FIVE-LEVEL HIERARCHY STRIP -------------------------------------- */}
         <section className="hierarchy-strip">
-          <div className="kicker">Six-level rollout hierarchy</div>
+          <div className="kicker">Five-level rollout hierarchy</div>
           <div className="hierarchy-row">
             {[
               {
@@ -175,7 +175,7 @@ export default async function HomePage() {
                 ex: "SWE-bench family, MMLU family",
               },
               {
-                name: "Suite",
+                name: "Composite",
                 count: formatNumber(compositeCount),
                 ex: "Open LLM Leaderboard v2, HELM Instruct",
               },
@@ -185,14 +185,9 @@ export default async function HomePage() {
                 ex: "GSM8K, IFEval, MMLU-Pro",
               },
               {
-                name: "Split",
+                name: "Slice",
                 count: formatNumber(sliceCount),
-                ex: "algebra (within MATH), Python (Multi-SWE-Bench)",
-              },
-              {
-                name: "Subtask",
-                count: "—",
-                ex: "level-5, multi-turn",
+                ex: "algebra (within MATH), level-5, multi-turn",
               },
               {
                 name: "Metric",

components/benchmark-detail.tsx

@@ -62,10 +62,10 @@ interface BenchmarkVariant {
   evaluation: BenchmarkEvaluation
   result: EvaluationResult
   label: string
-  variantType: "setup" | "subtask" | "setup+subtask" | "default"
+  variantType: "setup" | "slice" | "setup+slice" | "default"
   metricLabel: string
   setupLabel: string | null
-  subtaskLabel: string | null
+  sliceLabel: string | null
   displayScore: string
   normalizedScore: number
   rankPosition: number | null
@@ -92,9 +92,9 @@ interface BenchmarkGroup {
   variants: BenchmarkVariant[]
 }
 
-interface SuiteGroup {
-  suiteKey: string
-  suiteName: string
+interface CompositeGroup {
+  compositeKey: string
+  compositeName: string
   benchmarks: BenchmarkGroup[]
   avgRawScore: number
   avgNormalizedScore: number
@@ -269,77 +269,77 @@ function getSourceTypeDisplayName(value: string | null | undefined) {
   return normalizeDisplayLabel(value?.replace(/_/g, " ")) || "Unknown"
 }
 
-function normalizeSuiteKey(key: string): string {
+function normalizeCompositeKey(key: string): string {
   const k = key.toLowerCase().replace(/[-.\s]+/g, "_").replace(/^_+|_+$/g, "")
   if (/^fibble\d*_arena$/.test(k)) return "fibble_arena"
   if (/^arc_agi_v\d+/.test(k)) return "arc_agi"
   return k
 }
 
-function doesLabelMatchSuiteKey(label: string | null | undefined, suiteKey: string) {
+function doesLabelMatchSuiteKey(label: string | null | undefined, compositeKey: string) {
   if (!label) {
     return false
   }
 
-  return normalizeSuiteKey(normalizeDisplayKey(label)) === normalizeSuiteKey(suiteKey)
+  return normalizeCompositeKey(normalizeDisplayKey(label)) === normalizeCompositeKey(compositeKey)
 }
 
-function getSuiteKey(group: BenchmarkGroup): string {
+function getCompositeKey(group: BenchmarkGroup): string {
   const evaluation = group.variants[0]?.evaluation
   const backendSuiteKey =
     evaluation?.benchmark_parent_key ||
     evaluation?.benchmark_family_key ||
     evaluation?.benchmark
 
-  return normalizeSuiteKey(backendSuiteKey ?? group.key)
+  return normalizeCompositeKey(backendSuiteKey ?? group.key)
 }
 
-function getSuiteDisplayName(key: string): string {
-  const normalizedKey = normalizeSuiteKey(key)
+function getCompositeDisplayName(key: string): string {
+  const normalizedKey = normalizeCompositeKey(key)
   return SUITE_DISPLAY_NAMES[normalizedKey] ?? normalizeDisplayLabel(key)
 }
 
-function getSuiteName(group: BenchmarkGroup, suiteKey: string): string {
+function getCompositeName(group: BenchmarkGroup, compositeKey: string): string {
   const evaluation = group.variants[0]?.evaluation
   const benchmarkCardName = group.benchmarkCard?.benchmark_details?.name
   const backendParentName = evaluation?.benchmark_parent_name
   const backendFamilyName = evaluation?.benchmark_family_name
 
-  if (doesLabelMatchSuiteKey(backendParentName, suiteKey)) {
+  if (doesLabelMatchSuiteKey(backendParentName, compositeKey)) {
     return normalizeDisplayLabel(backendParentName)
   }
 
-  if (doesLabelMatchSuiteKey(backendFamilyName, suiteKey)) {
+  if (doesLabelMatchSuiteKey(backendFamilyName, compositeKey)) {
     return normalizeDisplayLabel(backendFamilyName)
   }
 
-  if (doesLabelMatchSuiteKey(benchmarkCardName, suiteKey)) {
+  if (doesLabelMatchSuiteKey(benchmarkCardName, compositeKey)) {
     return normalizeDisplayLabel(benchmarkCardName)
   }
 
-  return getSuiteDisplayName(suiteKey)
+  return getCompositeDisplayName(compositeKey)
 }
 
-function groupBySuite(
+function groupByComposite(
   groups: BenchmarkGroup[],
   modelIds: string[],
   peerRanks: PeerRanksMap
-): SuiteGroup[] {
-  const suites = new Map<string, BenchmarkGroup[]>()
+): CompositeGroup[] {
+  const composites = new Map<string, BenchmarkGroup[]>()
   for (const group of groups) {
-    const key = getSuiteKey(group)
-    const existing = suites.get(key) ?? []
+    const key = getCompositeKey(group)
+    const existing = composites.get(key) ?? []
     existing.push(group)
-    suites.set(key, existing)
+    composites.set(key, existing)
   }
 
-  return Array.from(suites.entries()).map(([suiteKey, benchmarks]) => {
+  return Array.from(composites.entries()).map(([compositeKey, benchmarks]) => {
     const scores = benchmarks.map(b => b.avgNormalizedScore).filter(Number.isFinite)
     const avgScore = scores.length > 0 ? scores.reduce((a, b) => a + b, 0) / scores.length : 0
     const rawScores = benchmarks.map((benchmark) => benchmark.avgRawScore).filter(Number.isFinite)
     const avgRawScore = rawScores.length > 0 ? rawScores.reduce((a, b) => a + b, 0) / rawScores.length : 0
 
-    // Find best rank across all benchmarks in suite
+    // Find best rank across all benchmarks in the composite
     let bestRank: { position: number; total: number } | null = null
     for (const b of benchmarks) {
       const rank = getGroupPeerRank(b, modelIds, peerRanks)
@@ -350,8 +350,8 @@ function groupBySuite(
     }
 
     return {
-      suiteKey,
-      suiteName: benchmarks[0] ? getSuiteName(benchmarks[0], suiteKey) : getSuiteDisplayName(suiteKey),
+      compositeKey,
+      compositeName: benchmarks[0] ? getCompositeName(benchmarks[0], compositeKey) : getCompositeDisplayName(compositeKey),
       benchmarks,
       avgRawScore,
       avgNormalizedScore: avgScore,
@@ -359,7 +359,7 @@ function groupBySuite(
       bestRank,
     }
   }).sort((a, b) => {
-    // Sort by best peer rank ratio (lower = better); unranked suites go to the bottom
+    // Sort by best peer rank ratio (lower = better); unranked composites go to the bottom
     const aRatio = a.bestRank ? a.bestRank.position / (a.bestRank.total || a.bestRank.position) : Infinity
     const bRatio = b.bestRank ? b.bestRank.position / (b.bestRank.total || b.bestRank.position) : Infinity
     if (aRatio !== bRatio) return aRatio - bRatio
@@ -466,25 +466,25 @@ function getMetricDisplayLabel(result: EvaluationResult) {
 function getVariantDescriptor(
   evaluation: BenchmarkEvaluation,
   result: EvaluationResult
-): Pick<BenchmarkVariant, "label" | "variantType" | "metricLabel" | "setupLabel" | "subtaskLabel"> {
+): Pick<BenchmarkVariant, "label" | "variantType" | "metricLabel" | "setupLabel" | "sliceLabel"> {
   const evaluationVariantRaw = getEvaluationVariantLabel(evaluation)
   const evaluationVariant = evaluationVariantRaw ? formatSetupDisplayLabel(evaluationVariantRaw) : null
   const metricLabel = getMetricDisplayLabel(result)
   const metricKey = normalizeDisplayKey(metricLabel)
   const metricIsAmbiguous = AMBIGUOUS_GROUP_LABELS.has(metricKey)
-  const subtaskLabel = evaluation.slice_name ? normalizeDisplayLabel(evaluation.slice_name) : null
+  const sliceLabel = evaluation.slice_name ? normalizeDisplayLabel(evaluation.slice_name) : null
   const setupLabel = evaluationVariant ? formatSetupDisplayLabel(evaluationVariant) : null
-  const baseLabel = subtaskLabel
-    ? (metricIsAmbiguous ? subtaskLabel : `${subtaskLabel} · ${metricLabel}`)
+  const baseLabel = sliceLabel
+    ? (metricIsAmbiguous ? sliceLabel : `${sliceLabel} · ${metricLabel}`)
     : metricLabel
 
-  if (setupLabel && subtaskLabel) {
+  if (setupLabel && sliceLabel) {
     return {
       label: `${setupLabel} · ${baseLabel}`,
-      variantType: "setup+subtask",
+      variantType: "setup+slice",
       metricLabel,
       setupLabel,
-      subtaskLabel,
+      sliceLabel,
     }
   }
 
@@ -494,17 +494,17 @@ function getVariantDescriptor(
       variantType: "setup",
       metricLabel,
       setupLabel,
-      subtaskLabel: null,
+      sliceLabel: null,
     }
   }
 
-  if (subtaskLabel || !metricIsAmbiguous) {
+  if (sliceLabel || !metricIsAmbiguous) {
     return {
       label: baseLabel,
-      variantType: subtaskLabel ? "subtask" : "default",
+      variantType: sliceLabel ? "slice" : "default",
       metricLabel,
       setupLabel: null,
-      subtaskLabel: subtaskLabel ?? null,
+      sliceLabel: sliceLabel ?? null,
     }
   }
 
@@ -513,7 +513,7 @@ function getVariantDescriptor(
     variantType: "default",
     metricLabel,
     setupLabel: null,
-    subtaskLabel: null,
+    sliceLabel: null,
   }
 }
 
@@ -603,7 +603,7 @@ function getTableConfigLabel(row: VariantRowData) {
     return row.variant.setupLabel
   }
 
-  if (row.variant.variantType === "subtask") {
+  if (row.variant.variantType === "slice") {
     return "Default setup"
   }
 
@@ -865,9 +865,9 @@ function getVariantTypeTone(variantType: BenchmarkVariant["variantType"]) {
   switch (variantType) {
     case "setup":
       return "bg-sky-100 text-sky-800 dark:bg-sky-950/50 dark:text-sky-300"
-    case "subtask":
+    case "slice":
       return "bg-violet-100 text-violet-800 dark:bg-violet-950/50 dark:text-violet-300"
-    case "setup+subtask":
+    case "setup+slice":
       return "bg-amber-100 text-amber-800 dark:bg-amber-950/50 dark:text-amber-300"
     default:
       return "bg-muted text-muted-foreground"
@@ -878,10 +878,10 @@ function getVariantTypeLabel(variantType: BenchmarkVariant["variantType"]) {
   switch (variantType) {
     case "setup":
       return "Setup change"
-    case "subtask":
-      return "Benchmark subtask"
-    case "setup+subtask":
-      return "Setup + subtask"
+    case "slice":
+      return "Benchmark slice"
+    case "setup+slice":
+      return "Setup + slice"
     default:
       return "Single run"
   }
@@ -911,8 +911,8 @@ function formatSetupDisplayLabel(setupLabel: string | null) {
 }
 
 function getVariantPrimaryLabel(variant: BenchmarkVariant, groupTitle: string) {
-  if (variant.subtaskLabel) {
-    return variant.subtaskLabel
+  if (variant.sliceLabel) {
+    return variant.sliceLabel
   }
 
   if (variant.metricLabel) {
@@ -929,41 +929,41 @@ function getVariantPrimaryLabel(variant: BenchmarkVariant, groupTitle: string) {
   return variant.label
 }
 
-function getGroupSubtaskLabels(group: BenchmarkGroup) {
+function getGroupSliceLabels(group: BenchmarkGroup) {
   return Array.from(
     new Set(
       group.variants
-        .map((variant) => variant.subtaskLabel?.trim())
+        .map((variant) => variant.sliceLabel?.trim())
         .filter((label): label is string => Boolean(label))
     )
   )
 }
 
-function getGroupSubtaskCount(group: BenchmarkGroup) {
-  return getGroupSubtaskLabels(group).length
+function getGroupSliceCount(group: BenchmarkGroup) {
+  return getGroupSliceLabels(group).length
 }
 
 function getBenchmarkGroupHeading(group: BenchmarkGroup) {
   return group.canonicalTitle
 }
 
-function getSuiteBadgeMeta(suite: SuiteGroup) {
-  if (suite.benchmarks.length > 1) {
+function getCompositeBadgeMeta(composite: CompositeGroup) {
+  if (composite.benchmarks.length > 1) {
     return {
-      count: suite.benchmarks.length,
-      label: `sub-benchmark${suite.benchmarks.length === 1 ? "" : "s"}`,
+      count: composite.benchmarks.length,
+      label: `sub-benchmark${composite.benchmarks.length === 1 ? "" : "s"}`,
       className:
         "border-sky-200/80 bg-sky-50/70 text-sky-700 dark:border-sky-900/60 dark:bg-sky-950/30 dark:text-sky-300",
     }
   }
 
-  const singleGroup = suite.benchmarks[0]
+  const singleGroup = composite.benchmarks[0]
   if (!singleGroup) {
     return null
   }
 
-  const suiteMatchesBenchmark = normalizeSuiteKey(suite.suiteName) === normalizeSuiteKey(singleGroup.title)
-  if (!suiteMatchesBenchmark) {
+  const compositeMatchesBenchmark = normalizeCompositeKey(composite.compositeName) === normalizeCompositeKey(singleGroup.title)
+  if (!compositeMatchesBenchmark) {
     return {
       count: 1,
       label: "sub-benchmark",
@@ -972,11 +972,11 @@ function getSuiteBadgeMeta(suite: SuiteGroup) {
     }
   }
 
-  const subtaskCount = getGroupSubtaskCount(singleGroup)
-  if (subtaskCount > 0) {
+  const sliceCount = getGroupSliceCount(singleGroup)
+  if (sliceCount > 0) {
     return {
-      count: subtaskCount,
-      label: `subtask${subtaskCount === 1 ? "" : "s"}`,
+      count: sliceCount,
+      label: `slice${sliceCount === 1 ? "" : "s"}`,
       className:
         "border-emerald-200/70 bg-emerald-50 text-emerald-700 dark:border-emerald-900/60 dark:bg-emerald-950/30 dark:text-emerald-300",
     }
@@ -1405,7 +1405,7 @@ function getVariantDedupKey(variant: BenchmarkVariant) {
     metricLabel: variant.metricLabel,
     variantType: variant.variantType,
     setupLabel: variant.setupLabel,
-    subtaskLabel: variant.subtaskLabel,
+    sliceLabel: variant.sliceLabel,
     displayScore: variant.displayScore,
     sourceOrganization: getOrganizationDisplayName(variant.evaluation.source_metadata.source_organization_name),
     sourceName: normalizeDisplayLabel(variant.evaluation.source_metadata.source_name ?? ""),
@@ -1459,7 +1459,7 @@ function buildBenchmarkGroups(
       variantType: descriptor.variantType,
       metricLabel: descriptor.metricLabel,
       setupLabel: descriptor.setupLabel,
-      subtaskLabel: descriptor.subtaskLabel,
+      sliceLabel: descriptor.sliceLabel,
       displayScore,
       normalizedScore,
       rankPosition,
@@ -1570,10 +1570,10 @@ function buildBenchmarkGroups(
 
       group.variants = Array.from(dedupedVariants.values())
       group.variants.sort((a, b) => {
-        const aIsSubtask = Boolean(a.evaluation.slice_key)
-        const bIsSubtask = Boolean(b.evaluation.slice_key)
-        if (aIsSubtask !== bIsSubtask) {
-          return aIsSubtask ? 1 : -1
+        const aIsSlice = Boolean(a.evaluation.slice_key)
+        const bIsSlice = Boolean(b.evaluation.slice_key)
+        if (aIsSlice !== bIsSlice) {
+          return aIsSlice ? 1 : -1
         }
 
         const aPrimaryLabel = getVariantPrimaryLabel(a, group.title)
@@ -2043,9 +2043,9 @@ export function BenchmarkDetail({
       }))
   }, [filteredBenchmarkGroups, comparisonIndex, summary.categories_covered])
 
-  const suiteGroups = useMemo(() => {
-    const groups = groupBySuite(filteredBenchmarkGroups, modelIds, peerRanks)
-    // Re-sort suites by max relevance of their benchmarks
+  const compositeGroups = useMemo(() => {
+    const groups = groupByComposite(filteredBenchmarkGroups, modelIds, peerRanks)
+    // Re-sort composites by max relevance of their benchmarks
     return groups.sort((a, b) => {
       const aMax = Math.max(...a.benchmarks.map(getRelevanceScore))
       const bMax = Math.max(...b.benchmarks.map(getRelevanceScore))
@@ -2053,48 +2053,48 @@ export function BenchmarkDetail({
     })
   }, [filteredBenchmarkGroups, modelIds, peerRanks, getRelevanceScore])
 
-  const categorySuiteSections = useMemo(
+  const categoryCompositeSections = useMemo(
     () =>
       groupedFilteredBenchmarkGroups
         .map(({ category, groups }) => ({
           category,
-          suites: groupBySuite(groups, modelIds, peerRanks).sort((a, b) => {
+          composites: groupByComposite(groups, modelIds, peerRanks).sort((a, b) => {
             const aMax = Math.max(...a.benchmarks.map(getRelevanceScore))
             const bMax = Math.max(...b.benchmarks.map(getRelevanceScore))
             return bMax - aMax
           }),
         }))
-        .filter((section) => section.suites.length > 0),
+        .filter((section) => section.composites.length > 0),
     [groupedFilteredBenchmarkGroups, modelIds, peerRanks, getRelevanceScore]
   )
 
   const categoryScoreRanges = useMemo(() => {
     const ranges = new Map<CategoryType, ScoreRange>()
 
-    for (const section of categorySuiteSections) {
+    for (const section of categoryCompositeSections) {
       ranges.set(
         section.category,
-        getScoreRange(section.suites.map((suite) => suite.avgNormalizedScore))
+        getScoreRange(section.composites.map((composite) => composite.avgNormalizedScore))
       )
     }
 
     return ranges
-  }, [categorySuiteSections])
+  }, [categoryCompositeSections])
 
-  const suiteBenchmarkScoreRanges = useMemo(() => {
+  const compositeBenchmarkScoreRanges = useMemo(() => {
     const ranges = new Map<string, ScoreRange>()
 
-    for (const section of categorySuiteSections) {
-      for (const suite of section.suites) {
+    for (const section of categoryCompositeSections) {
+      for (const composite of section.composites) {
         ranges.set(
-          suite.suiteKey,
-          getScoreRange(suite.benchmarks.map((group) => group.avgNormalizedScore))
+          composite.compositeKey,
+          getScoreRange(composite.benchmarks.map((group) => group.avgNormalizedScore))
         )
       }
     }
 
     return ranges
-  }, [categorySuiteSections])
+  }, [categoryCompositeSections])
 
   const benchmarkGroupLookup = useMemo(
     () => new Map(benchmarkGroups.map((group) => [group.key, group] as const)),
@@ -2104,11 +2104,11 @@ export function BenchmarkDetail({
     ? benchmarkGroupLookup.get(activeBenchmarkGroupKey) ?? null
     : null
 
-  const toggleSuite = (suiteKey: string) => {
+  const toggleSuite = (compositeKey: string) => {
     setExpandedSuites((prev) => {
       const next = new Set(prev)
-      if (next.has(suiteKey)) next.delete(suiteKey)
-      else next.add(suiteKey)
+      if (next.has(compositeKey)) next.delete(compositeKey)
+      else next.add(compositeKey)
       return next
     })
   }
@@ -2150,10 +2150,10 @@ export function BenchmarkDetail({
   )
   const repeatedBenchmarkCount = overviewBenchmarkGroups.filter((group) => group.variants.length > 1).length
   const setupDrivenBenchmarkCount = overviewBenchmarkGroups.filter((group) =>
-    group.variants.some((variant) => variant.variantType === "setup" || variant.variantType === "setup+subtask")
+    group.variants.some((variant) => variant.variantType === "setup" || variant.variantType === "setup+slice")
   ).length
-  const subtaskDrivenBenchmarkCount = overviewBenchmarkGroups.filter((group) =>
-    group.variants.some((variant) => variant.variantType === "subtask" || variant.variantType === "setup+subtask")
+  const sliceDrivenBenchmarkCount = overviewBenchmarkGroups.filter((group) =>
+    group.variants.some((variant) => variant.variantType === "slice" || variant.variantType === "setup+slice")
   ).length
 
   useEffect(() => {
@@ -2414,7 +2414,7 @@ export function BenchmarkDetail({
     summary.model_info.name,
   ])
 
-  // A plotbox can expose a top-level "view" selector (subtasks, child
+  // A plotbox can expose a top-level "view" selector (slices, child
   // benchmarks, components) and an optional metric tab rail beneath the chart.
   // Plotbox grouping is driven entirely by comparison-index's own
   // benchmark_family_key so it stays in sync with the backend.
@@ -2448,7 +2448,7 @@ export function BenchmarkDetail({
     familyName: string
     category: CategoryType
     kind: "single-eval" | "multi-eval"
-    childKindLabel: "metric" | "benchmark" | "component" | "subtask" | null
+    childKindLabel: "metric" | "benchmark" | "component" | "slice" | null
     views: PlotboxView[]
     primaryGroup: BenchmarkGroup
   }
@@ -2534,7 +2534,7 @@ export function BenchmarkDetail({
       const { familyName, category, resolved } = bucket
 
       if (resolved.length === 1) {
-        // One eval in scope — subtasks/splits become the view selector while
+        // One eval in scope — slices/splits become the view selector while
         // metrics move to a compact tab rail beneath the chart.
         const { group, evalEntry } = resolved[0]
         const evalDisplay =
@@ -2545,10 +2545,10 @@ export function BenchmarkDetail({
         >()
 
         for (const variant of group.variants) {
-          const viewKey = variant.subtaskLabel
-            ? `subtask:${normalizeDisplayKey(variant.subtaskLabel)}`
+          const viewKey = variant.sliceLabel
+            ? `slice:${normalizeDisplayKey(variant.sliceLabel)}`
             : "default"
-          const label = variant.subtaskLabel || "Overall"
+          const label = variant.sliceLabel || "Overall"
           const bucketForView = singleEvalViewBuckets.get(viewKey) ?? {
             viewKey,
             label,
@@ -2600,7 +2600,7 @@ export function BenchmarkDetail({
           familyName: evalDisplay,
           category,
           kind: "single-eval",
-          childKindLabel: views.length > 1 ? "subtask" : null,
+          childKindLabel: views.length > 1 ? "slice" : null,
           views,
           primaryGroup: group,
         })
@@ -2656,7 +2656,7 @@ export function BenchmarkDetail({
       if (views.length === 0) continue
 
       let hasComponent = false
-      let hasSubtask = false
+      let hasSlice = false
       let hasDistinctLeaves = false
       for (const r of children) {
         const leafKey = r.evalEntry.benchmark_leaf_key
@@ -2666,17 +2666,17 @@ export function BenchmarkDetail({
         if (r.group.variants[0]?.evaluation.benchmark_component_key ?? null) {
           hasComponent = true
         } else {
-          hasSubtask = true
+          hasSlice = true
         }
       }
       const childKindLabel: PlotboxUnit["childKindLabel"] =
-        hasComponent && hasSubtask
+        hasComponent && hasSlice
           ? "component"
           : hasComponent
             ? "metric"
             : hasDistinctLeaves
               ? "benchmark"
-              : "subtask"
+              : "slice"
 
       units.push({
         unitKey: `family:${famKey}`,
@@ -2909,8 +2909,8 @@ export function BenchmarkDetail({
     const childKindPlural =
       unit.childKindLabel === "metric"
         ? childKindCount === 1 ? "metric" : "metrics"
-        : unit.childKindLabel === "subtask"
-          ? childKindCount === 1 ? "subtask" : "subtasks"
+        : unit.childKindLabel === "slice"
+          ? childKindCount === 1 ? "slice" : "slices"
           : unit.childKindLabel === "benchmark"
             ? childKindCount === 1 ? "benchmark" : "benchmarks"
             : childKindCount === 1 ? "component" : "components"
@@ -3395,10 +3395,10 @@ export function BenchmarkDetail({
                 ? `${reportingStats.missingGenerationConfigs} entries are missing generation config, limiting cross-slice comparability.`
                 : "Generation configuration is present across the result set."}
             </p>
-            {(setupDrivenBenchmarkCount > 0 || subtaskDrivenBenchmarkCount > 0) && (
+            {(setupDrivenBenchmarkCount > 0 || sliceDrivenBenchmarkCount > 0) && (
               <p className="text-[13px] leading-[1.7] text-[color:var(--fg-muted)]">
                 Decomposition: <span className="text-[color:var(--fg)]">{setupDrivenBenchmarkCount}</span> setup-aware ·{" "}
-                <span className="text-[color:var(--fg)]">{subtaskDrivenBenchmarkCount}</span> subtask-aware.
+                <span className="text-[color:var(--fg)]">{sliceDrivenBenchmarkCount}</span> slice-aware.
                 {reportingStats.libraryList.length > 0 && (
                   <>
                     {" "}Eval libraries: <span className="text-[color:var(--fg)]">{reportingStats.libraryList.join(", ")}</span>.
@@ -3531,7 +3531,7 @@ export function BenchmarkDetail({
 
         <p className="text-[14px] leading-[1.7] text-[color:var(--fg-muted)] max-w-[64rem] mb-6">
           {isResearchView
-            ? "Benchmark-first view of this model's reported results, grouped by category. Setup spread and subtask-vs-setup differences surface up-front."
+            ? "Benchmark-first view of this model's reported results, grouped by category. Setup spread and slice-vs-setup differences surface up-front."
             : "The public evidence behind this model, grouped by category. The strongest and most variable signals are listed first."}
           {policyHighlights.length > 0 && !isResearchView && (
             <>
@@ -3606,9 +3606,9 @@ export function BenchmarkDetail({
             )}
             {repeatedBenchmarkCount > 0 && (
               <>
-                <dt>Subtask spread</dt>
+                <dt>Slice spread</dt>
                 <dd>
-                  {repeatedBenchmarkCount} benchmark{repeatedBenchmarkCount === 1 ? "" : "s"} include multiple subtasks or setups.
+                  {repeatedBenchmarkCount} benchmark{repeatedBenchmarkCount === 1 ? "" : "s"} include multiple slices or setups.
                 </dd>
               </>
             )}
@@ -3701,7 +3701,7 @@ export function BenchmarkDetail({
             <option value="rank">Sort · Best rank</option>
             <option value="score">Sort · Highest score</option>
             <option value="name">Sort · Name (A–Z)</option>
-            <option value="variants">Sort · Most subtasks</option>
+            <option value="variants">Sort · Most slices</option>
             <option value="spread">Sort · Largest spread</option>
           </select>
         </div>
@@ -3856,7 +3856,7 @@ export function BenchmarkDetail({
                             <span>{row.variant.result.generation_config.num_few_shot}-shot</span>
                           )}
                           {row.variant.setupLabel && <span>· {row.variant.setupLabel}</span>}
-                          {row.variant.subtaskLabel && <span>· {row.variant.subtaskLabel}</span>}
+                          {row.variant.sliceLabel && <span>· {row.variant.sliceLabel}</span>}
                         </div>
                       )}
                     </div>
@@ -4527,22 +4527,22 @@ function BenchmarkResultCard({
                     <Separator className="my-4" />
                     <div className="mb-2">
                       <div className="text-xs font-semibold text-muted-foreground uppercase tracking-wider">Detailed Breakdown</div>
-                      <div className="text-xs text-muted-foreground mt-1">Scores and structured metadata for individual subtasks or metrics</div>
+                      <div className="text-xs text-muted-foreground mt-1">Scores and structured metadata for individual slices or metrics</div>
                     </div>
 
                     {numericBreakdown.length > 0 && (
                       <div className="grid grid-cols-1 sm:grid-cols-2 xl:grid-cols-3 gap-3">
                         {numericBreakdown.map(([key, value]) => {
                         let valDisplay = typeof value === 'number' ? value.toFixed(2) : value;
-                        let normalized_subtask = 0;
+                        let normalized_slice = 0;
                         
                         if (typeof value === 'number') {
                             if (unit === 'accuracy' || !unit || unit === 'pass@1') {
                                 valDisplay = formatRawScoreValue(value);
-                                normalized_subtask = value;
+                                normalized_slice = value;
                             } else {
                                 valDisplay = value.toFixed(2);
-                                normalized_subtask = (value - min_score) / (max_score - min_score);
+                                normalized_slice = (value - min_score) / (max_score - min_score);
                             }
                         }
                         
@@ -4556,7 +4556,7 @@ function BenchmarkResultCard({
                             {valDisplay}
                           </div>
                           {typeof value === 'number' && (
-                            <Progress value={normalized_subtask * 100} className="h-1 mt-2" />
+                            <Progress value={normalized_slice * 100} className="h-1 mt-2" />
                           )}
                         </div>
                       )})}
@@ -4800,7 +4800,7 @@ function AggregatedBenchmarkCard({
     Number.isFinite(latestTimestamp) ? formatCompactDate(String(latestTimestamp)) : formatCompactDate(group.variants[0]?.evaluation.retrieved_timestamp ?? "")
   const compactDomains = group.domains.slice(0, 2)
   const progressWidth = Math.max(4, Math.min(100, group.avgNormalizedScore * 100))
-  const subtaskCount = getGroupSubtaskCount(group)
+  const sliceCount = getGroupSliceCount(group)
 
   const toggleRow = (rowKey: string) => {
     setExpandedRows((current) => ({
@@ -4851,9 +4851,9 @@ function AggregatedBenchmarkCard({
                       card
                     </span>
                   )}
-                  {subtaskCount > 0 && (
+                  {sliceCount > 0 && (
                     <span className="shrink-0 rounded-full border border-emerald-200/70 bg-emerald-50 px-1.5 py-0.5 text-[9px] font-semibold uppercase tracking-[0.12em] text-emerald-700 dark:border-emerald-900/60 dark:bg-emerald-950/30 dark:text-emerald-300">
-                      {subtaskCount} subtask{subtaskCount === 1 ? "" : "s"}
+                      {sliceCount} slice{sliceCount === 1 ? "" : "s"}
                     </span>
                   )}
                   {compactDomains.map((domain) => (
@@ -4881,7 +4881,7 @@ function AggregatedBenchmarkCard({
                 )}
               </div>
 
-              {/* Subtask count */}
+              {/* Slice count */}
               <span className="shrink-0 text-[11px] text-muted-foreground w-16 text-right hidden sm:block">
                 {group.variants.length} {group.variants.length === 1 ? "row" : "rows"}
               </span>
@@ -5077,9 +5077,9 @@ function AggregatedBenchmarkCard({
                                   </div>
                                   <div className="text-xs text-muted-foreground">
                                     {variant.setupLabel && <span>Setup: {variant.setupLabel}</span>}
-                                    {variant.setupLabel && variant.subtaskLabel && <span> • </span>}
-                                    {variant.subtaskLabel && <span>Subtask: {variant.subtaskLabel}</span>}
-                                    {!variant.setupLabel && !variant.subtaskLabel && <span>{group.title}</span>}
+                                    {variant.setupLabel && variant.sliceLabel && <span> • </span>}
+                                    {variant.sliceLabel && <span>Slice: {variant.sliceLabel}</span>}
+                                    {!variant.setupLabel && !variant.sliceLabel && <span>{group.title}</span>}
                                   </div>
                                 </div>
                               </div>
@@ -5193,8 +5193,8 @@ function BenchmarkDeepDiveDialogPanel({
   const [resolvedRanks, setResolvedRanks] = useState<Record<string, { position: number; total: number | null }>>({})
   const [isResolvingRanks, setIsResolvingRanks] = useState(false)
   const compactDomains = group.domains.slice(0, 2)
-  const subtaskCount = getGroupSubtaskCount(group)
-  const hasSubtaskMatrix = subtaskCount > 0
+  const sliceCount = getGroupSliceCount(group)
+  const hasSliceMatrix = sliceCount > 0
   const sourceOrganizations = useMemo(
     () => new Set(group.variants.map((variant) => getOrganizationDisplayName(variant.evaluation.source_metadata.source_organization_name))),
     [group.variants]
@@ -5335,18 +5335,18 @@ function BenchmarkDeepDiveDialogPanel({
   // Kept only to drive the single-setup overview: when every reported row is
   // reported under the same setup, the detail table collapses into a compact
   // view that drops the redundant "Reporting setup" column.
-  const subtaskSetups = useMemo(() => {
-    if (!hasSubtaskMatrix) return null
+  const sliceSetups = useMemo(() => {
+    if (!hasSliceMatrix) return null
     const setupOrder: string[] = []
     for (const row of variantRows) {
       const setupDisplayLabel = formatSetupDisplayLabel(row.variant.setupLabel)
       if (!setupOrder.includes(setupDisplayLabel)) setupOrder.push(setupDisplayLabel)
     }
     return { setupOrder }
-  }, [hasSubtaskMatrix, variantRows])
+  }, [hasSliceMatrix, variantRows])
 
-  const useSingleSetupOverview = Boolean(subtaskSetups && subtaskSetups.setupOrder.length === 1)
-  const singleSetupDisplayLabel = useSingleSetupOverview ? subtaskSetups?.setupOrder[0] ?? null : null
+  const useSingleSetupOverview = Boolean(sliceSetups && sliceSetups.setupOrder.length === 1)
+  const singleSetupDisplayLabel = useSingleSetupOverview ? sliceSetups?.setupOrder[0] ?? null : null
 
   useEffect(() => {
     const pendingRows = variantRows.filter(
@@ -5411,7 +5411,7 @@ function BenchmarkDeepDiveDialogPanel({
             </DialogTitle>
             <DialogDescription className="mt-1.5 text-[13px] leading-[1.5] text-[color:var(--fg-muted)]">
               {isResearchView
-                ? "Inspect setup subtasks, score details, and source provenance in one focused view."
+                ? "Inspect setup slices, score details, and source provenance in one focused view."
                 : "Inspect reporting setup and evidence details before interpreting benchmark position."}
             </DialogDescription>
             {(compactDomains.length > 0 || group.benchmarkCard) && (
@@ -5462,9 +5462,9 @@ function BenchmarkDeepDiveDialogPanel({
             <div className="kicker">Sources</div>
             <div className="mt-1 text-[18px] font-semibold tabular-nums">
               {sourceOrganizations.size}
-              {hasSubtaskMatrix && (
+              {hasSliceMatrix && (
                 <span className="ml-2 font-mono text-[10px] tracking-[0.12em] uppercase text-[color:var(--fg-subtle)]">
-                  · {subtaskCount} subtask{subtaskCount === 1 ? "" : "s"}
+                  · {sliceCount} slice{sliceCount === 1 ? "" : "s"}
                 </span>
               )}
             </div>
@@ -5482,7 +5482,7 @@ function BenchmarkDeepDiveDialogPanel({
 
         {/* Sources — distinct reporting orgs and dataset links for this group.
             Pulled up to the top of the deep-dive so the per-row table can
-            stay focused on subtask / setup / score. */}
+            stay focused on slice / setup / score. */}
         {(() => {
           type SourceEntry = {
             key: string
@@ -5581,7 +5581,7 @@ function BenchmarkDeepDiveDialogPanel({
         {useSingleSetupOverview ? (
           <section>
             <div className="section-head">
-              <h2>{hasAmbiguousPrimaryLabels ? "Reported runs" : "Subtask overview"}</h2>
+              <h2>{hasAmbiguousPrimaryLabels ? "Reported runs" : "Slice overview"}</h2>
               <span className="font-mono text-[10px] uppercase tracking-[0.15em] text-[color:var(--fg-subtle)]">
                 {singleSetupDisplayLabel ? `${singleSetupDisplayLabel} · ` : ""}
                 {variantRows.length} row{variantRows.length === 1 ? "" : "s"}
@@ -5594,8 +5594,8 @@ function BenchmarkDeepDiveDialogPanel({
                   ? "These rows share the same benchmark label, so run names or differing config fields are surfaced to show what changed across reports."
                   : "These rows describe the same benchmark view, so the table surfaces the reported run name or setup differences that separate them."
                 : isResearchView
-                  ? "This benchmark reports one setup, so subtasks, scores, and provenance are merged into one comparison view."
-                  : "This benchmark only reports one setup, so the subtask evidence is consolidated into a single reader-friendly view."}
+                  ? "This benchmark reports one setup, so slices, scores, and provenance are merged into one comparison view."
+                  : "This benchmark only reports one setup, so the slice evidence is consolidated into a single reader-friendly view."}
             </p>
 
             <div className="min-h-0 overflow-auto">
@@ -5603,7 +5603,7 @@ function BenchmarkDeepDiveDialogPanel({
                 <thead>
                   <tr>
                     <th className="w-[60%]">
-                      {hasAmbiguousPrimaryLabels ? "Reported row" : "Subtask"}
+                      {hasAmbiguousPrimaryLabels ? "Reported row" : "Slice"}
                     </th>
                     <th className="w-[20%]">Setup detail</th>
                     <th className="num w-[10%]">Score</th>
@@ -5678,14 +5678,14 @@ function BenchmarkDeepDiveDialogPanel({
             </span>
           </div>
           <p className="mb-4 max-w-[60rem] text-[13px] leading-[1.65] text-[color:var(--fg-muted)]">
-            Primary row labels show the benchmark slice or subtask. Setup and source details sit alongside each row.
+            Primary row labels show the benchmark slice or slice. Setup and source details sit alongside each row.
           </p>
 
           <div className="min-h-0 overflow-auto">
             <table className="ec-htable table-fixed">
               <thead>
                 <tr>
-                  <th className="w-[46%]">Subtask</th>
+                  <th className="w-[46%]">Slice</th>
                   <th className="w-[36%]">Reporting setup</th>
                   <th className="num w-[9%]">Score</th>
                   <th className="num w-[9%]">Rank</th>
@@ -5895,7 +5895,7 @@ function VariantExpandedDetail({
             )}
             {sourceData?.dataset_version && <InlineMeta label="Dataset Version" value={sourceData.dataset_version} />}
             {sourceData?.hf_split && <InlineMeta label="Split" value={sourceData.hf_split} />}
-            {variant.subtaskLabel && <InlineMeta label="Subtask" value={normalizeDisplayLabel(variant.subtaskLabel)} />}
+            {variant.sliceLabel && <InlineMeta label="Slice" value={normalizeDisplayLabel(variant.sliceLabel)} />}
             {variant.setupLabel && <InlineMeta label="Setup" value={formatSetupDisplayLabel(variant.setupLabel)} />}
             {inferencePlatform && <InlineMeta label="Inference Platform" value={inferencePlatform} />}
             {variant.evaluation.source_metadata.source_name && (
@@ -5952,7 +5952,7 @@ function VariantExpandedDetail({
       {numericBreakdown.length > 0 && (
         <div className="space-y-3">
           <div className="text-xs font-semibold uppercase tracking-[0.2em] text-muted-foreground">
-            {isResearchView ? "Subtask Scores" : "Reported Metrics"}
+            {isResearchView ? "Slice Scores" : "Reported Metrics"}
           </div>
           <div className="grid gap-3 md:grid-cols-2 2xl:grid-cols-3">
             {numericBreakdown.map(([key, value]) => {

components/benchmark-evaluation-card.tsx

@@ -135,7 +135,7 @@ function formatScoreValue(value: number | null | undefined) {
 
 function getCoverageSummaryLabel(data: BenchmarkEvaluationCardData) {
   if (data.benchmarks_count > 0) {
-    return `${data.benchmarks_count} benchmark suite${data.benchmarks_count === 1 ? "" : "s"} surfaced`
+    return `${data.benchmarks_count} benchmark composite${data.benchmarks_count === 1 ? "" : "s"} surfaced`
   }
 
   if (data.latest_source_name) {
@@ -302,7 +302,7 @@ export function BenchmarkEvaluationCard({
                 <Badge variant="secondary">{data.variant_count} versions</Badge>
               )}
               {paramsBillions && <Badge variant="secondary">{paramsBillions} parameters</Badge>}
-              <Badge variant="outline">{data.benchmarks_count} benchmark suites</Badge>
+              <Badge variant="outline">{data.benchmarks_count} benchmark composites</Badge>
               <Badge variant="outline">{data.evaluations_count} reported results</Badge>
               {reproducibilityGapCount > 0 && (
                 <Badge

components/eval-card.tsx

@@ -120,7 +120,7 @@ export function EvalCard({ summary, delayMs = 0 }: EvalCardProps) {
         <div className="min-w-0">
           <div className="text-xl font-bold">{summary.evaluation_name}</div>
           <div className="mt-1 text-xs font-medium uppercase tracking-[0.18em] text-muted-foreground">
-            Suite: {summary.composite_benchmark_name}
+            Composite: {summary.composite_benchmark_name}
           </div>
           <div className="mt-1 text-sm text-muted-foreground line-clamp-2">
             {/*

components/eval-detail.tsx

@@ -113,14 +113,14 @@ interface SliceTab {
  * - <= SLICE_PILL_THRESHOLD: render every slice as a pill (current familiar UX).
  * - > SLICE_PILL_THRESHOLD: render "All slices" + currently-selected pill +
  *   a "Browse N slices" button that opens a searchable dialog. Hundreds of
- *   subtasks (e.g. AIRBench's 374) fit cleanly.
+ *   slices (e.g. AIRBench's 374) fit cleanly.
  */
 function SliceSelector({
-  activeSubtaskTab,
+  activeSliceTab,
   onChange,
   tabs,
 }: {
-  activeSubtaskTab: string
+  activeSliceTab: string
   onChange: (key: string) => void
   tabs: SliceTab[]
 }) {
@@ -128,7 +128,7 @@ function SliceSelector({
   const [search, setSearch] = useState("")
 
   const useBrowser = tabs.length > SLICE_PILL_THRESHOLD
-  const activeTab = tabs.find((tab) => tab.key === activeSubtaskTab)
+  const activeTab = tabs.find((tab) => tab.key === activeSliceTab)
 
   const filteredTabs = useMemo(() => {
     const query = search.trim().toLowerCase()
@@ -143,7 +143,7 @@ function SliceSelector({
         <div className="flex flex-wrap gap-2">
           <button
             type="button"
-            className={`ec-pill${activeSubtaskTab === "all" ? " on" : ""}`}
+            className={`ec-pill${activeSliceTab === "all" ? " on" : ""}`}
             onClick={() => onChange("all")}
           >
             All slices
@@ -152,7 +152,7 @@ function SliceSelector({
             <button
               key={tab.key}
               type="button"
-              className={`ec-pill${activeSubtaskTab === tab.key ? " on" : ""}`}
+              className={`ec-pill${activeSliceTab === tab.key ? " on" : ""}`}
               onClick={() => onChange(tab.key)}
             >
               {tab.label}
@@ -172,7 +172,7 @@ function SliceSelector({
       <div className="flex flex-wrap items-center gap-2">
         <button
           type="button"
-          className={`ec-pill${activeSubtaskTab === "all" ? " on" : ""}`}
+          className={`ec-pill${activeSliceTab === "all" ? " on" : ""}`}
           onClick={() => onChange("all")}
         >
           All slices
@@ -230,11 +230,11 @@ function SliceSelector({
               }}
               className={cn(
                 "flex w-full items-center justify-between border-b px-4 py-2.5 text-left text-sm transition-colors hover:bg-muted/40",
-                activeSubtaskTab === "all" && "bg-muted/40 font-semibold"
+                activeSliceTab === "all" && "bg-muted/40 font-semibold"
               )}
             >
               <span>All slices (no filter)</span>
-              {activeSubtaskTab === "all" && <span className="text-xs text-muted-foreground">selected</span>}
+              {activeSliceTab === "all" && <span className="text-xs text-muted-foreground">selected</span>}
             </button>
             {filteredTabs.length === 0 ? (
               <div className="px-4 py-6 text-center text-sm text-muted-foreground">
@@ -251,11 +251,11 @@ function SliceSelector({
                   }}
                   className={cn(
                     "flex w-full items-center justify-between border-b px-4 py-2 text-left text-sm transition-colors hover:bg-muted/40 last:border-b-0",
-                    activeSubtaskTab === tab.key && "bg-muted/40 font-semibold"
+                    activeSliceTab === tab.key && "bg-muted/40 font-semibold"
                   )}
                 >
                   <span className="min-w-0 truncate pr-2">{tab.label}</span>
-                  {activeSubtaskTab === tab.key && (
+                  {activeSliceTab === tab.key && (
                     <span className="shrink-0 text-xs text-muted-foreground">selected</span>
                   )}
                 </button>
@@ -543,10 +543,10 @@ export function EvalDetail({ summary }: EvalDetailProps) {
     : "Not linked"
   const leaderboardDescription = isResearchView
     ? summary.is_aggregated
-      ? "Models ranked by average raw score across the suite's component benchmarks."
+      ? "Models ranked by average raw score across the composite's component benchmarks."
       : "Models ranked by raw score for this benchmark."
     : summary.is_aggregated
-      ? "Averaged model results across the suite's component benchmarks, with drill-down to each component score."
+      ? "Averaged model results across the composite's component benchmarks, with drill-down to each component score."
       : "Model results with benchmark context, source dataset detail, and optional instance-data links."
   const reportingCompleteness = summary.evalcards?.annotations?.reporting_completeness
   const documentationPopulatedCount = reportingCompleteness
@@ -656,7 +656,7 @@ export function EvalDetail({ summary }: EvalDetailProps) {
                 className="font-mono text-[10px] uppercase tracking-[0.12em]"
                 style={{ color: "var(--fg-subtle)" }}
               >
-                metric spec · completeness · comparability{summary.subtasks?.length ? " · subtasks" : ""}
+                metric spec · completeness · comparability{summary.subtasks?.length ? " · slices" : ""}
               </span>
             </div>
             {overviewOpen ? (
@@ -678,10 +678,10 @@ export function EvalDetail({ summary }: EvalDetailProps) {
                 {isResearchView ? "Metric specification" : "Reading context"}
               </div>
               <dl className="ec-datalist">
-                <dt>Suite</dt>
+                <dt>Composite</dt>
                 <dd>
                   {summary.is_aggregated
-                    ? summary.aggregate_sources?.map((source) => source.composite_benchmark_name).join(", ") || "Multiple suites"
+                    ? summary.aggregate_sources?.map((source) => source.composite_benchmark_name).join(", ") || "Multiple composites"
                     : summary.composite_benchmark_name}
                 </dd>
                 <dt>{isResearchView ? "Benchmark ID" : "What this covers"}</dt>
@@ -744,7 +744,7 @@ export function EvalDetail({ summary }: EvalDetailProps) {
               >
                 <div className="kicker mb-2">Benchmark structure</div>
                 <p className="text-[13px] mb-4" style={{ color: "var(--fg-muted)", maxWidth: 640 }}>
-                  Benchmark-level summary metrics and subtask slices grouped in one compact section.
+                  Benchmark-level summary metrics and slices grouped in one compact section.
                 </p>
 
                 {summary.root_metrics && summary.root_metrics.length > 0 && (
@@ -776,15 +776,15 @@ export function EvalDetail({ summary }: EvalDetailProps) {
                       className="font-mono uppercase mb-1"
                       style={{ fontSize: 10, letterSpacing: "0.14em", color: "var(--fg-subtle)" }}
                     >
-                      Subtask breakdown · {summary.subtasks.length}
+                      Slice breakdown · {summary.subtasks.length}
                     </div>
                     <ul
                       className="flex flex-col"
                       style={{ borderTop: "1px solid var(--border-soft)" }}
                     >
-                      {summary.subtasks.map((subtask) => (
+                      {summary.subtasks.map((slice) => (
                         <li
-                          key={subtask.subtask_key}
+                          key={slice.subtask_key}
                           className="grid gap-x-4 py-3"
                           style={{
                             gridTemplateColumns: "minmax(160px, 280px) 1fr",
@@ -793,20 +793,20 @@ export function EvalDetail({ summary }: EvalDetailProps) {
                         >
                           <div className="min-w-0">
                             <div className="font-semibold text-[13px] truncate">
-                              {subtask.display_name || subtask.subtask_name}
+                              {slice.display_name || slice.subtask_name}
                             </div>
-                            {subtask.canonical_display_name && subtask.canonical_display_name !== (subtask.display_name || subtask.subtask_name) && (
+                            {slice.canonical_display_name && slice.canonical_display_name !== (slice.display_name || slice.subtask_name) && (
                               <div
                                 className="mt-0.5 font-mono text-[10px] uppercase tracking-[0.08em] truncate"
                                 style={{ color: "var(--fg-subtle)" }}
-                                title={subtask.canonical_display_name}
+                                title={slice.canonical_display_name}
                               >
-                                {subtask.canonical_display_name}
+                                {slice.canonical_display_name}
                               </div>
                             )}
                           </div>
                           <div className="flex flex-wrap items-center gap-1.5">
-                            {subtask.metrics.map((metric) => (
+                            {slice.metrics.map((metric) => (
                               <span
                                 key={metric.metric_summary_id}
                                 className="ec-tag"
@@ -919,13 +919,13 @@ export function EvalDetail({ summary }: EvalDetailProps) {
               <tbody>
                 {pagedLeaderboardRows.map(({ key, rank, modelResult, normalizedScore }) => {
                   const isExpanded = expandedRows[key] ?? false
-                  const subtasks = modelResult.score_details.details
+                  const slices = modelResult.score_details.details
                     ? Object.entries(modelResult.score_details.details).filter(([, value]) => typeof value === "number")
                     : []
                   const hasExpandableDetails =
                     isResearchView ||
                     (modelResult.aggregate_components && modelResult.aggregate_components.length > 1) ||
-                    subtasks.length > 1
+                    slices.length > 1
 
                   const datasetName = Array.isArray(modelResult.source_data)
                     ? undefined
@@ -1256,28 +1256,28 @@ export function EvalDetail({ summary }: EvalDetailProps) {
                                 </div>
                               )}
 
-                              {subtasks.length > 1 && (
+                              {slices.length > 1 && (
                                 <div className="space-y-2">
                                   <div
                                     className="font-mono uppercase"
                                     style={{ fontSize: 10, letterSpacing: "0.14em", color: "var(--fg-subtle)" }}
                                   >
-                                    Subtask breakdown
+                                    Slice breakdown
                                   </div>
                                   <div className="overflow-x-auto" style={{ border: "1px solid var(--border-soft)" }}>
                                     <table className="ec-htable">
                                       <thead>
                                         <tr>
-                                          <th>Subtask</th>
+                                          <th>Slice</th>
                                           <th className="num">Raw</th>
                                         </tr>
                                       </thead>
                                       <tbody>
-                                        {subtasks.map(([subtaskName, value]) => {
+                                        {slices.map(([sliceName, value]) => {
                                           const numericValue = value as number
                                           return (
-                                            <tr key={subtaskName}>
-                                              <td className="font-medium text-[13px] capitalize">{subtaskName.replace(/_/g, " ")}</td>
+                                            <tr key={sliceName}>
+                                              <td className="font-medium text-[13px] capitalize">{sliceName.replace(/_/g, " ")}</td>
                                               <td className="num font-mono tabular-nums text-[13px]" style={{ color: "var(--fg-muted)" }}>
                                                 {formatRawScore(numericValue, summary.metric_config.unit)}
                                               </td>
@@ -1453,7 +1453,7 @@ function MultiMetricLeaderboard({
     return root?.column_key ?? metrics[0]?.column_key ?? "model"
   })
   const [sortDirection, setSortDirection] = useState<"asc" | "desc">("desc")
-  const [activeSubtaskTab, setActiveSubtaskTab] = useState<string>("all")
+  const [activeSliceTab, setActiveSliceTab] = useState<string>("all")
   const [minParamStep, setMinParamStep] = useState(0)
   const [maxParamStep, setMaxParamStep] = useState(PARAM_RANGE_MAX_INDEX)
   const [expandedRows, setExpandedRows] = useState<Record<string, boolean>>({})
@@ -1486,7 +1486,7 @@ function MultiMetricLeaderboard({
   const leaderboardRows = summary.leaderboard_rows ?? []
   const allMetricKeys = useMemo(() => leaderboardMetrics.map((metric) => metric.column_key), [leaderboardMetrics])
   // Cap default visible columns to avoid hangs on benchmarks with hundreds of metrics
-  // (e.g. helm_air_bench has 374 subtask×metric pairs). Users can opt in to more.
+  // (e.g. helm_air_bench has 374 slice×metric pairs). Users can opt in to more.
   const DEFAULT_VISIBLE_METRIC_CAP = 24
   const defaultVisibleMetricKeys = useMemo(
     () => allMetricKeys.slice(0, DEFAULT_VISIBLE_METRIC_CAP),
@@ -1498,7 +1498,7 @@ function MultiMetricLeaderboard({
     [leaderboardMetrics]
   )
   const visibleMetricKeySet = useMemo(() => new Set(visibleMetricKeys), [visibleMetricKeys])
-  const subtaskMetricCounts = useMemo(() => {
+  const sliceMetricCounts = useMemo(() => {
     const counts = new Map<string, number>()
     for (const metric of leaderboardMetrics) {
       if (metric.scope === "subtask" && metric.subtask_key) {
@@ -1508,16 +1508,16 @@ function MultiMetricLeaderboard({
     return counts
   }, [leaderboardMetrics])
 
-  const singleMetricSubtaskTabs = useMemo(() => {
+  const singleMetricSliceTabs = useMemo(() => {
     return leaderboardMetrics
-      .filter((metric) => metric.scope === "subtask" && metric.subtask_key && subtaskMetricCounts.get(metric.subtask_key) === 1)
+      .filter((metric) => metric.scope === "subtask" && metric.subtask_key && sliceMetricCounts.get(metric.subtask_key) === 1)
       .map((metric) => ({
         key: metric.subtask_key as string,
         label: metric.subtask_name ?? getCompactMetricLabel(metric.display_name),
       }))
-  }, [leaderboardMetrics, subtaskMetricCounts])
+  }, [leaderboardMetrics, sliceMetricCounts])
 
-  const hasSubtaskTabs = singleMetricSubtaskTabs.length > 1
+  const hasSliceTabs = singleMetricSliceTabs.length > 1
 
   const visibleMetrics = useMemo(
     () =>
@@ -1526,13 +1526,13 @@ function MultiMetricLeaderboard({
           return false
         }
 
-        if (!hasSubtaskTabs || activeSubtaskTab === "all") {
+        if (!hasSliceTabs || activeSliceTab === "all") {
           return true
         }
 
-        return metric.scope === "subtask" && metric.subtask_key === activeSubtaskTab
+        return metric.scope === "subtask" && metric.subtask_key === activeSliceTab
       }),
-    [activeSubtaskTab, hasSubtaskTabs, leaderboardMetrics, visibleMetricKeySet]
+    [activeSliceTab, hasSliceTabs, leaderboardMetrics, visibleMetricKeySet]
   )
   const visibleMetricColumnKeySet = useMemo(
     () => new Set(visibleMetrics.map((metric) => metric.column_key)),
@@ -1628,7 +1628,7 @@ function MultiMetricLeaderboard({
   }, [defaultVisibleMetricKeys, summary.evaluation_id])
 
   useEffect(() => {
-    setActiveSubtaskTab("all")
+    setActiveSliceTab("all")
   }, [summary.evaluation_id])
 
   useEffect(() => {
@@ -1648,21 +1648,21 @@ function MultiMetricLeaderboard({
   }, [leaderboardMetricMap, leaderboardMetrics, sortKey, visibleMetricColumnKeySet])
 
   useEffect(() => {
-    if (!hasSubtaskTabs) {
-      if (activeSubtaskTab !== "all") {
-        setActiveSubtaskTab("all")
+    if (!hasSliceTabs) {
+      if (activeSliceTab !== "all") {
+        setActiveSliceTab("all")
       }
       return
     }
 
-    if (activeSubtaskTab === "all") {
+    if (activeSliceTab === "all") {
       return
     }
 
-    if (!singleMetricSubtaskTabs.some((tab) => tab.key === activeSubtaskTab)) {
-      setActiveSubtaskTab("all")
+    if (!singleMetricSliceTabs.some((tab) => tab.key === activeSliceTab)) {
+      setActiveSliceTab("all")
     }
-  }, [activeSubtaskTab, hasSubtaskTabs, singleMetricSubtaskTabs])
+  }, [activeSliceTab, hasSliceTabs, singleMetricSliceTabs])
 
   const pagedRows = useMemo(
     () => sortedRows.slice(0, page * 50),
@@ -1761,7 +1761,7 @@ function MultiMetricLeaderboard({
             {leaderboardMetrics.map((metric) => {
               const isVisible = visibleMetricKeySet.has(metric.column_key)
               const isLastVisible = isVisible && visibleMetrics.length === 1
-              const visibleLabel = metric.scope === "subtask" && metric.subtask_key && subtaskMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
+              const visibleLabel = metric.scope === "subtask" && metric.subtask_key && sliceMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
                 ? metric.subtask_name
                 : getCompactMetricLabel(metric.display_name)
 
@@ -1793,7 +1793,7 @@ function MultiMetricLeaderboard({
               .filter((v): v is number => isNumericScore(v))
             if (values.length < 3) return null
             const label =
-              metric.scope === "subtask" && metric.subtask_key && subtaskMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
+              metric.scope === "subtask" && metric.subtask_key && sliceMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
                 ? metric.subtask_name
                 : getCompactMetricLabel(metric.display_name)
             return {
@@ -1816,12 +1816,12 @@ function MultiMetricLeaderboard({
       })()}
 
       <div className="ec-card" style={{ padding: 0, overflow: "hidden" }}>
-        {hasSubtaskTabs && (
+        {hasSliceTabs && (
           <div className="border-b bg-background px-5 py-3 sm:px-6">
             <SliceSelector
-              activeSubtaskTab={activeSubtaskTab}
-              onChange={setActiveSubtaskTab}
-              tabs={singleMetricSubtaskTabs}
+              activeSliceTab={activeSliceTab}
+              onChange={setActiveSliceTab}
+              tabs={singleMetricSliceTabs}
             />
           </div>
         )}
@@ -1868,13 +1868,13 @@ function MultiMetricLeaderboard({
                   {getSortIndicator("developer")}
                 </th>
                 {visibleMetrics.map((metric) => {
-                  const showSubtaskTopline =
-                    !hasSubtaskTabs &&
-                    !(metric.scope === "subtask" && metric.subtask_key && subtaskMetricCounts.get(metric.subtask_key) === 1) &&
+                  const showSliceTopline =
+                    !hasSliceTabs &&
+                    !(metric.scope === "subtask" && metric.subtask_key && sliceMetricCounts.get(metric.subtask_key) === 1) &&
                     metric.scope === "subtask" &&
                     metric.subtask_name
                   const mainLabel =
-                    metric.scope === "subtask" && metric.subtask_key && subtaskMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
+                    metric.scope === "subtask" && metric.subtask_key && sliceMetricCounts.get(metric.subtask_key) === 1 && metric.subtask_name
                       ? metric.subtask_name
                       : getCompactMetricLabel(metric.display_name)
                   return (
@@ -1885,7 +1885,7 @@ function MultiMetricLeaderboard({
                       onClick={() => handleSort(metric.column_key)}
                       title={describeLeaderboardMetric(metric)}
                     >
-                      {showSubtaskTopline && (
+                      {showSliceTopline && (
                         <div
                           className="font-mono normal-case"
                           style={{

components/family-table.tsx

@@ -4,7 +4,7 @@ import { Fragment, useMemo, useState } from "react"
 import { useRouter } from "next/navigation"
 import { ArrowUpRight, ChevronDown, ChevronRight } from "lucide-react"
 
-import type { HierarchyFamily, HierarchyLeaf } from "@/lib/backend-artifacts"
+import type { HierarchyBenchmark, HierarchyFamily, HierarchyLeaf } from "@/lib/backend-artifacts"
 import type { BenchmarkCard, CategoryType } from "@/lib/benchmark-schema"
 import type { BenchmarkEvalListItem } from "@/lib/eval-processing"
 
@@ -67,18 +67,60 @@ interface LeafEntry {
   domains: string[]
 }
 
+/**
+ * Build the per-row list of expandable benchmarks under a family.
+ *
+ * Primary path: the v2 production shape, where the hierarchy nests
+ * benchmarks under `families[].composites[].benchmarks[]` (plus
+ * `standalone_benchmarks[]` and the family-level `benchmarks[]`). Each
+ * benchmark carries its own `summary_eval_ids` for navigation.
+ *
+ * Fallback path: the legacy `families[].leaves[]` shape used by older
+ * snapshots. Kept for compatibility while older caches are still in
+ * circulation.
+ */
 function collectLeafEntries(
   fam: HierarchyFamily,
   benchmarkCards?: Record<string, BenchmarkCard>,
 ): LeafEntry[] {
   const out: LeafEntry[] = []
+
+  // ── Primary: nested benchmarks (v2) ────────────────────────────────
+  const nested: HierarchyBenchmark[] = [
+    ...(fam.standalone_benchmarks ?? []),
+    ...(fam.benchmarks ?? []),
+    ...(fam.composites ?? []).flatMap((c) => c.benchmarks ?? []),
+  ]
+  for (const benchmark of nested) {
+    const summaryIds = benchmark.summary_eval_ids ?? []
+    const ids =
+      summaryIds.length > 0
+        ? summaryIds
+        : benchmark.key
+        ? [`${fam.key}_${benchmark.key}`, benchmark.key]
+        : []
+    if (ids.length === 0) continue
+    const collected = new Set<string>()
+    for (const d of benchmark.tags?.domains ?? []) collected.add(d.toLowerCase())
+    const cardByKey = benchmarkCards?.[benchmark.key]
+    for (const d of cardByKey?.benchmark_details?.domains ?? []) collected.add(d.toLowerCase())
+    for (const id of ids) {
+      const cardById = benchmarkCards?.[id]
+      for (const d of cardById?.benchmark_details?.domains ?? []) collected.add(d.toLowerCase())
+    }
+    out.push({
+      id: ids[0],
+      leafKey: benchmark.key,
+      leafName: benchmark.display_name || benchmark.key,
+      evalsCount: ids.length,
+      domains: Array.from(collected),
+    })
+  }
+
+  if (out.length > 0) return out
+
+  // ── Fallback: legacy `leaves` shape ────────────────────────────────
   for (const leaf of fam.leaves ?? []) {
-    // Backends differ in whether leaves carry an explicit
-    // `eval_summary_ids` array. When absent, fall back to the
-    // pipeline's standard `${fam.key}_${leaf.key}` naming, then to the
-    // bare leaf key — both are stable enough for the detail page to
-    // resolve. This stops the inline benchmarks grid from disappearing
-    // on a backend that ships hierarchy.json without leaf eval ids.
     const explicit = leaf.eval_summary_ids ?? []
     const ids =
       explicit.length > 0
@@ -87,10 +129,6 @@ function collectLeafEntries(
         ? [`${fam.key}_${leaf.key}`, leaf.key]
         : []
     if (ids.length === 0) continue
-    // Domain sources, in order of trust:
-    //   (1) hierarchy `leaf.tags.domains` — sometimes absent
-    //   (2) benchmark-metadata keyed by leaf.key
-    //   (3) benchmark-metadata keyed by the leaf's eval_summary_id
     const collected = new Set<string>()
     for (const d of leaf.tags?.domains ?? []) collected.add(d.toLowerCase())
     const cardByLeaf = benchmarkCards?.[leaf.key]
@@ -107,6 +145,7 @@ function collectLeafEntries(
       domains: Array.from(collected),
     })
   }
+
   return out
 }
 

components/model-compare-dialog.tsx

@@ -428,7 +428,7 @@ export function ModelCompareDialog({
                                 ) : null}
                                 {row.key === "latest" ? (
                                   <div className="flex items-center gap-2">
-                                    <span>{model.latest_source_name || `${model.benchmarks_count} benchmark suites summarized`}</span>
+                                    <span>{model.latest_source_name || `${model.benchmarks_count} benchmark composites summarized`}</span>
                                     {model.source_urls[0] ? (
                                       <a
                                         href={model.source_urls[0]}

components/policy-overview.tsx

@@ -51,7 +51,7 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
   const card = summary.benchmark_card
 
   // Defensive check: pipelines older than the "ancestor card leak" fix
-  // sometimes attach the parent suite's card to a leaf benchmark (e.g.
+  // sometimes attach the parent composite's card to a leaf benchmark (e.g.
   // helm_classic's card embedded under XSUM). Detect when the card's own
   // name is clearly not this benchmark and ignore its narrative text — the
   // synthesized fallback below produces something accurate instead.
@@ -81,17 +81,20 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
   const goal = cardMatchesEval ? card?.purpose_and_intended_users?.goal?.trim() || "" : ""
 
   // Detect "parent" benchmark pages — either an aggregated composite or a
-  // multi-metric matrix where each column is a subtask. In both cases the
+  // multi-metric matrix where each column is a slice. In both cases the
   // per-evaluation `metric_config.evaluation_description` belongs to whichever
   // component was processed first (e.g. just the "airline" subset of Tau
   // Bench 2) and would mislead a policy reader. Synthesize parent framing
-  // instead and surface the subtasks separately.
+  // instead and surface the slices separately.
+  // Note: at the data layer the backend still ships these as `subtasks` /
+  // `metric.scope === "subtask"`; we read those fields directly but label
+  // them "slice" in the UI.
   const isAggregated = summary.is_aggregated === true
   const aggregateNames = (summary.aggregate_sources ?? [])
     .map((s) => s.composite_benchmark_name)
     .filter((s): s is string => typeof s === "string" && s.length > 0)
 
-  const subtaskLabels = useMemo(() => {
+  const sliceLabels = useMemo(() => {
     const seen = new Set<string>()
     const labels: string[] = []
     const add = (raw: string | undefined | null) => {
@@ -103,8 +106,8 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
       seen.add(key)
       labels.push(trimmed)
     }
-    for (const subtask of summary.subtasks ?? []) {
-      add(subtask.display_name || subtask.subtask_name)
+    for (const slice of summary.subtasks ?? []) {
+      add(slice.display_name || slice.subtask_name)
     }
     for (const metric of summary.leaderboard_metrics ?? []) {
       if (metric.scope === "subtask") {
@@ -116,14 +119,14 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
   }, [summary.subtasks, summary.leaderboard_metrics, aggregateNames])
 
   const isMatrix = (summary.leaderboard_metrics?.length ?? 0) > 1
-  const isParentPage = isAggregated || (isMatrix && subtaskLabels.length > 1)
+  const isParentPage = isAggregated || (isMatrix && sliceLabels.length > 1)
   const useComponentDescription = !isParentPage
 
-  const parentFallback = isParentPage && subtaskLabels.length > 1
-    ? `${summary.evaluation_name} reports results across ${subtaskLabels.length} ${
-        isAggregated ? "component benchmarks" : "subtasks"
+  const parentFallback = isParentPage && sliceLabels.length > 1
+    ? `${summary.evaluation_name} reports results across ${sliceLabels.length} ${
+        isAggregated ? "component benchmarks" : "slices"
       }. Each is evaluated separately; the score shown is the ${
-        isAggregated ? "average" : "per-subtask result"
+        isAggregated ? "average" : "per-slice result"
       }.`
     : null
 
@@ -134,7 +137,7 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
     (useComponentDescription ? summary.metric_config.evaluation_description : summary.evaluation_name)
 
   const [expanded, setExpanded] = useState(false)
-  const [subtasksOpen, setSubtasksOpen] = useState(false)
+  const [slicesOpen, setSlicesOpen] = useState(false)
   const isLong = summaryText.length > SUMMARY_PREVIEW_CHARS
   const visibleText = expanded || !isLong
     ? summaryText
@@ -299,15 +302,15 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
         </dd>
       </dl>
 
-      {isParentPage && subtaskLabels.length > 1 && (
+      {isParentPage && sliceLabels.length > 1 && (
         <div
           className="mt-4"
           style={{ border: "1px solid var(--border-soft)", background: "var(--bg)" }}
         >
           <button
             type="button"
-            onClick={() => setSubtasksOpen((v) => !v)}
-            aria-expanded={subtasksOpen}
+            onClick={() => setSlicesOpen((v) => !v)}
+            aria-expanded={slicesOpen}
             className="flex w-full items-center justify-between gap-3 px-3.5 py-2.5 text-left transition-colors hover:bg-[color:var(--bg-warm)]"
           >
             <span
@@ -315,20 +318,20 @@ export function PolicyOverview({ summary }: PolicyOverviewProps) {
               style={{ fontSize: 10, letterSpacing: "0.14em", color: "var(--fg-muted)" }}
             >
               <Layers className="h-3.5 w-3.5" />
-              {isAggregated ? `Component benchmarks (${subtaskLabels.length})` : `Subtasks (${subtaskLabels.length})`}
+              {isAggregated ? `Component benchmarks (${sliceLabels.length})` : `Slices (${sliceLabels.length})`}
             </span>
-            {subtasksOpen ? (
+            {slicesOpen ? (
               <ChevronUp className="h-4 w-4" style={{ color: "var(--fg-muted)" }} />
             ) : (
               <ChevronDown className="h-4 w-4" style={{ color: "var(--fg-muted)" }} />
             )}
           </button>
-          {subtasksOpen && (
+          {slicesOpen && (
             <ul
               className="grid list-disc gap-x-6 gap-y-1 px-3.5 pb-3.5 pl-9 text-[13px] sm:grid-cols-2 lg:grid-cols-3"
               style={{ color: "var(--fg)" }}
             >
-              {subtaskLabels.map((name) => (
+              {sliceLabels.map((name) => (
                 <li key={name} className="capitalize">
                   {name}
                 </li>