compare: live-data emission floor, noindex thin ad hoc pairs, brand casing

src/app/compare/[slug]/page.tsx

@@ -198,6 +198,40 @@ export async function generateMetadata({
   const sharedCount = sharedSlugsForMeta.filter((s) => !excluded.has(s)).length;
   const benchWord = sharedCount === 1 ? "benchmark" : "benchmarks";
 
+  // Thin-content gate (SEO audit 2026-07-08): a pair whose shared
+  // benches carry live data for both providers on fewer than 2 of them
+  // renders either "awaiting live measurements" or a single card.
+  // Those pages stay reachable (internal links + stale index entries
+  // must not 404) but are marked noindex so direct hits stop counting
+  // against the domain. Mirrors the >= 2 live-shared emission floor in
+  // src/lib/compare/adhoc-pairs.ts so the sitemap never advertises a
+  // noindexed URL. Live rule matches liveResults(): not "unavailable"
+  // and p50 > 0, read off the already-loaded appearances.
+  const aLive = new Set(
+    a.appearances
+      .filter(
+        (x) => x.result.availability !== "unavailable" && x.result.ms.p50 > 0,
+      )
+      .map((x) => x.benchmark.slug),
+  );
+  const bLive = new Set(
+    b.appearances
+      .filter(
+        (x) => x.result.availability !== "unavailable" && x.result.ms.p50 > 0,
+      )
+      .map((x) => x.benchmark.slug),
+  );
+  const liveSharedCount = sharedSlugsForMeta.filter(
+    (s) => !excluded.has(s) && aLive.has(s) && bLive.has(s),
+  ).length;
+  // Curated pairs (explicit benchmarks list in COMPARE_PAIRS) are
+  // hand-picked head-term targets like usdc-vs-usdt and carry editorial
+  // framing beyond the ledger, so they stay indexable even with a
+  // single live shared bench. The gate only applies to combinatorial
+  // ad hoc pairs.
+  const isCurated = !!pair.benchmarks;
+  const thin = !isCurated && liveSharedCount < 2;
+
   // Meta description: unique per pair via the shared-count + provider
   // names + date. Kills the identical duplicate-content signal that had
   // Bing indexing 2 of 4938 compare pages. Also cites "as of DATE" for
@@ -211,6 +245,9 @@ export async function generateMetadata({
   return {
     title,
     description,
+    // follow stays on so PageRank keeps flowing through the body links
+    // (both provider pages, parent benches) even while deindexed.
+    ...(thin ? { robots: { index: false, follow: true } } : {}),
     alternates: { canonical: url },
     openGraph: {
       title,

src/app/compare/page.tsx

@@ -49,11 +49,20 @@ export default async function ComparePage() {
   // enumeration as sitemap.ts (shared lib), minus the curated pairs
   // already rendered above.
   const curatedSlugs = new Set(pairs.map((p) => p.slug));
-  const morePairs = adHocPairs(await getProviders())
+  const profiles = await getProviders();
+  // Prefer the profile display name (carries the bench spec's brand
+  // casing: dRPC, 1RPC, USDC) over the canonicalize() title-case
+  // fallback, which rendered "Drpc vs Tenderly" style labels here.
+  const nameBySlug = new Map(profiles.map((p) => [p.slug, p.name]));
+  const displayName = (slug: string): string => {
+    const canon = canonicalize(slug);
+    return nameBySlug.get(canon.slug) ?? canon.name;
+  };
+  const morePairs = adHocPairs(profiles)
     .filter((p) => !curatedSlugs.has(p.slug))
     .map((p) => ({
       ...p,
-      label: `${canonicalize(p.a).name} vs ${canonicalize(p.b).name}`,
+      label: `${displayName(p.a)} vs ${displayName(p.b)}`,
     }))
     .sort((a, b) => a.label.localeCompare(b.label));
 
@@ -64,7 +73,7 @@ export default async function ComparePage() {
     itemListElement: pairs.map((pair, i) => ({
       "@type": "ListItem",
       position: i + 1,
-      name: `${canonicalize(pair.providerA).name} vs ${canonicalize(pair.providerB).name}`,
+      name: `${displayName(pair.providerA)} vs ${displayName(pair.providerB)}`,
       url: `${SITE.url}/compare/${pair.slug}`,
     })),
   };
@@ -101,8 +110,14 @@ export default async function ComparePage() {
         </p>
         <ul className="mt-8 grid grid-cols-1 sm:grid-cols-2 gap-3">
           {pairs.map((pair) => {
-            const a = canonicalize(pair.providerA);
-            const b = canonicalize(pair.providerB);
+            const a = {
+              slug: canonicalize(pair.providerA).slug,
+              name: displayName(pair.providerA),
+            };
+            const b = {
+              slug: canonicalize(pair.providerB).slug,
+              name: displayName(pair.providerB),
+            };
             return (
               <li key={pair.slug}>
                 <Link

src/app/sitemap.ts

@@ -371,19 +371,21 @@ async function buildFullSitemap(): Promise<MetadataRoute.Sitemap> {
     lastModByPairSlug.set(pair.slug, pairLastMod(pair.providerA, pair.providerB));
   }
 
-  // Ad-hoc pair generation with hybrid threshold (SEO audit 2026-07-05):
+  // Ad-hoc pair generation with hybrid threshold (SEO audit 2026-07-08,
+  // tightened from 2026-07-05):
   //
-  //   - Both providers in BRAND_WHITELIST → emit at ≥ 1 shared bench.
-  //   - Otherwise → emit only at ≥ 3 shared benches.
+  //   - Both providers in BRAND_WHITELIST → emit at ≥ 2 shared benches
+  //     with live data for both.
+  //   - Otherwise → emit only at ≥ 3 live shared benches.
   //
-  // Previous rule (≥ 1 for any pair) generated 4938 ad-hoc URLs, 97% of
-  // which were near-duplicate templates over obscure providers. Bing
-  // indexed 2 pages out of 5093 and penalised the whole domain via
-  // thin-content signal. The hybrid keeps every commercial "X vs Y"
-  // pair a user might actually search for (helius-vs-mobula,
-  // alchemy-vs-moralis, chain-vs-chain, perp-vs-perp — 223 pairs) and
-  // adds only genuinely rich non-brand pairs (3 with ≥ 3 shared).
-  // Total: 226 ad-hoc + 21 curated ≈ 247 URLs.
+  // History: the original ≥ 1 for any pair generated 4938 ad-hoc URLs,
+  // 97% near-duplicate templates; Bing indexed 2 of 5093 and penalised
+  // the domain. The 07-05 hybrid (brand ≥ 1 structural) still emitted
+  // ~70 single-shared-bench pairs, many rendering "awaiting live
+  // measurements" (~2 KB). The ≥ 2 live floor matches the render-time
+  // noindex gate on /compare/[slug], so no sitemap URL is noindexed.
+  // Dropped pairs still render via the ad-hoc resolver (no 404s from
+  // bench-page or product-page cross links).
   // Enumeration shared with /compare (src/lib/compare/adhoc-pairs.ts)
   // so every advertised pair URL is also internally linked — sitemap-
   // only pairs were flagged as orphan pages (Ahrefs, 2026-07-08).

src/lib/compare-pairing.ts

@@ -83,7 +83,8 @@ async function buildCompareGraphRaw(): Promise<CompareGraph> {
  *  new provider or appearance flows in within minutes of being live. */
 const buildCompareGraphCached = unstable_cache(
   buildCompareGraphRaw,
-  ["compare-graph-v1"],
+  // v2: provider names switched to spec brand casing (providers v6).
+  ["compare-graph-v2"],
   { revalidate: 300, tags: ["benchmarks"] },
 );
 

src/lib/compare/adhoc-pairs.ts

@@ -11,34 +11,54 @@ const HEX_SLUG_RE = /^0x[0-9a-f]{4,}$/i;
  * sitemap-only pairs (e.g. binance-vs-tenderly) as orphan pages when
  * the two generators drifted.
  *
- * Hybrid threshold (SEO audit 2026-07-05): both providers in
- * BRAND_WHITELIST → emit at >= 1 shared bench; otherwise >= 3. See the
- * sitemap for the thin-content rationale (Bing penalty, 2026-07-05).
+ * Hybrid threshold (SEO audit 2026-07-08, tightened from 2026-07-05):
+ * both providers in BRAND_WHITELIST → emit at >= 2 shared benches with
+ * LIVE data for both; otherwise >= 3. "Live" mirrors liveResults():
+ * availability !== "unavailable" and p50 > 0, read straight off the
+ * already-loaded appearance results (no extra Prom round trip).
+ *
+ * Rationale for the bump from >= 1 structural to >= 2 live: the old
+ * rule emitted ~70 pairs whose single shared bench often had no live
+ * data ("awaiting live measurements", ~2 KB body). The >= 2 live floor
+ * also matches the render-time noindex gate on /compare/[slug] (pairs
+ * resolving < 2 benches with values are noindexed), so the sitemap
+ * never advertises a URL the page itself marks noindex. Tradeoff: a
+ * brand pair with exactly one live shared bench (real data, one card)
+ * leaves the sitemap and internal link lists too. It still renders at
+ * its URL via the ad-hoc resolver, so nothing 404s; it just stops
+ * being advertised until a second shared bench goes live.
  */
 export type AdHocPair = { a: string; b: string; slug: string };
 
 export function adHocPairs(profiles: ProviderProfile[]): AdHocPair[] {
-  const benchesBySlug = new Map<string, Set<string>>();
+  // bench slug → live flag, per provider. A shared bench only counts
+  // toward the threshold when it is live for BOTH providers.
+  const liveBenchesBySlug = new Map<string, Set<string>>();
   for (const p of profiles) {
     if (HEX_SLUG_RE.test(p.slug)) continue;
     if (CHAIN_BY_SLUG.has(p.slug)) continue;
-    const benches = new Set(p.appearances.map((a) => a.benchmark.slug));
-    if (benches.size >= 1) benchesBySlug.set(p.slug, benches);
+    const live = new Set<string>();
+    for (const a of p.appearances) {
+      if (a.result.availability === "unavailable") continue;
+      if (a.result.ms.p50 <= 0) continue;
+      live.add(a.benchmark.slug);
+    }
+    if (live.size >= 1) liveBenchesBySlug.set(p.slug, live);
   }
 
   const out: AdHocPair[] = [];
-  const slugList = [...benchesBySlug.keys()].sort();
+  const slugList = [...liveBenchesBySlug.keys()].sort();
   for (let i = 0; i < slugList.length; i += 1) {
     const aSlug = slugList[i];
-    const aBenches = benchesBySlug.get(aSlug)!;
+    const aBenches = liveBenchesBySlug.get(aSlug)!;
     for (let j = i + 1; j < slugList.length; j += 1) {
       const bSlug = slugList[j];
-      const bBenches = benchesBySlug.get(bSlug)!;
-      let shared = 0;
-      for (const s of aBenches) if (bBenches.has(s)) shared += 1;
+      const bBenches = liveBenchesBySlug.get(bSlug)!;
+      let sharedLive = 0;
+      for (const s of aBenches) if (bBenches.has(s)) sharedLive += 1;
       const bothBrand = BRAND_WHITELIST.has(aSlug) && BRAND_WHITELIST.has(bSlug);
-      const threshold = bothBrand ? 1 : 3;
-      if (shared < threshold) continue;
+      const threshold = bothBrand ? 2 : 3;
+      if (sharedLive < threshold) continue;
       out.push({ a: aSlug, b: bSlug, slug: `${aSlug}-vs-${bSlug}` });
     }
   }

src/lib/providers.ts

@@ -87,6 +87,31 @@ const CANONICAL_NAMES: Record<string, string> = {
   // brand casing.
   gram: "Gram",
   tonapi: "TonAPI",
+  // Brand casings the title-case fallback butchers ("Drpc", "Usdc",
+  // "Meowrpc" — SEO audit 2026-07-08). Profiles built from bench specs
+  // now inherit the spec's provider `name` field, but canonicalize() is
+  // also called with a bare slug (compare hub labels, OG images, search
+  // index) where no spec name is in scope, so the map stays the source
+  // of truth for these. Casing matches the bench spec `name` fields.
+  drpc: "dRPC",
+  "1rpc": "1RPC",
+  meowrpc: "MeowRPC",
+  usdc: "USDC",
+  usdt: "USDT",
+  dai: "DAI",
+  usde: "USDe",
+  fdusd: "FDUSD",
+  dydx: "dYdX",
+  okx: "OKX",
+  gmx: "GMX",
+  grvt: "GRVT",
+  cctp: "CCTP",
+  lifi: "LI.FI",
+  debridge: "deBridge",
+  zksync: "zkSync Era",
+  "near-intents": "NEAR Intents",
+  edgex: "edgeX",
+  sui: "Sui",
 };
 
 function titleCaseSlug(s: string): string {
@@ -103,6 +128,25 @@ export function canonicalize(slug: string): { slug: string; name: string } {
   return { slug: canon, name: CANONICAL_NAMES[canon] ?? titleCaseSlug(canon) };
 }
 
+/** Display name for a profile, in priority order:
+ *   1. CANONICAL_NAMES override (editorial, wins over everything).
+ *   2. The bench spec's provider `name` field — specs carry proper
+ *      brand casing (dRPC, 1RPC, USDC, dYdX v4) while the old
+ *      title-case fallback rendered "Drpc" / "Usdc" in every title,
+ *      H1, breadcrumb, and OG surface (SEO audit 2026-07-08). Only
+ *      trusted when the result's own slug IS the canonical slug; an
+ *      aliased sub-product (helius-sender) must not rename its parent.
+ *   3. Title-cased slug, last resort. */
+function profileDisplayName(
+  canonSlug: string,
+  r: ProviderResult,
+): string {
+  const override = CANONICAL_NAMES[canonSlug];
+  if (override) return override;
+  if (r.slug.toLowerCase() === canonSlug && r.name) return r.name;
+  return titleCaseSlug(canonSlug);
+}
+
 export type ProviderAppearance = {
   benchmark: Pick<
     Benchmark,
@@ -329,6 +373,13 @@ async function buildProviders(): Promise<ProviderProfile[]> {
         }
         existing.wins += winsEarned;
         if (!existing.type && r.type) existing.type = r.type;
+        // Upgrade a title-case fallback name the moment any appearance
+        // supplies a proper spec name (first appearance can come from
+        // an alias, which never carries the canonical brand casing).
+        if (existing.name === titleCaseSlug(key)) {
+          const better = profileDisplayName(key, r);
+          if (better !== existing.name) existing.name = better;
+        }
         if (chainsLed.length > 0) {
           existing.chainWins = existing.chainWins ?? {};
           for (const c of chainsLed) {
@@ -338,7 +389,7 @@ async function buildProviders(): Promise<ProviderProfile[]> {
       } else {
         byKey.set(key, {
           slug: canon.slug,
-          name: canon.name,
+          name: profileDisplayName(canon.slug, r),
           type: r.type,
           appearances: [appearance],
           wins: winsEarned,
@@ -448,7 +499,7 @@ function isBlacklistedSlug(slug: string): boolean {
 const PERP_VENUE_SEED = [
   { slug: "drift", name: "Drift" },
   { slug: "vertex", name: "Vertex" },
-  { slug: "edgex", name: "EdgeX" },
+  { slug: "edgex", name: "edgeX" },
   { slug: "extended", name: "Extended" },
   { slug: "aevo", name: "Aevo" },
   { slug: "pacifica", name: "Pacifica" },
@@ -482,7 +533,10 @@ const buildProvidersCached = unstable_cache(
   // newly identified HL frontends (invo, bitget-wallet, defi-saver,
   // unitywallet, marsgo, metamask-alt). Without the bump, /products/<slug>
   // 404s for them even after the YAML + builders.json updates landed.
-  ["providers-v5"],
+  // v6: profile names now inherit bench spec casing (dRPC, USDC, dYdX)
+  // instead of title-cased slugs. Bump flushes stale "Drpc"/"Usdc"
+  // names from every title/H1/breadcrumb surface.
+  ["providers-v6"],
   { revalidate: 60, tags: ["benchmarks"] },
 );