cherry-pick seo hygiene P0 → main

next.config.ts

@@ -121,6 +121,31 @@ const nextConfig: NextConfig = {
           },
         ],
       },
+      {
+        // Sitemap: force-dynamic in the route file (Data Cache 2MB cap
+        // blows past for the 500+ URL corpus), so nothing caches it by
+        // default and every crawler hit re-runs the full loader chain.
+        // Hold it edge-side for an hour with SWR so cold hits stay fast.
+        source: "/sitemap.xml",
+        headers: [
+          {
+            key: "Cache-Control",
+            value: "public, s-maxage=3600, stale-while-revalidate=86400",
+          },
+        ],
+      },
+      {
+        // Citable JSON is polled by LLM crawlers (Perplexity, ChatGPT,
+        // Claude Deep Research) — a bare `public` with no s-maxage sent
+        // every scrape to origin. Same freshness window as sitemap.
+        source: "/api/citable",
+        headers: [
+          {
+            key: "Cache-Control",
+            value: "public, s-maxage=3600, stale-while-revalidate=86400",
+          },
+        ],
+      },
     ];
   },
   async redirects() {

src/app/benchmarks/[slug]/opengraph-image.tsx

@@ -19,38 +19,17 @@ export async function generateStaticParams() {
   return [];
 }
 
-// Emit one OG image per (slug, chain) combo so social shares of
-// `/benchmarks/{slug}?chain=X` render a chain-honest card instead of the
-// cross-chain aggregate leader (which is the misleading baseline-skew
-// case bench-001 ran into: GMGN looking like "fastest crypto data API"
-// on the unfiltered view because Solana's faster baseline drags the
-// average down). The `id` Next.js threads through to the default image
-// handler is the chain value; the unfiltered card uses `id="default"`.
-export async function generateImageMetadata({
-  params,
-}: {
-  params: Promise<{ slug: string }>;
-}) {
-  const { slug } = await params;
-  // Load editorial-only via getBenchmark (memoised; safe at build time).
-  const b = await getBenchmark(slug);
-  const chains = (b?.dimensions?.chain ?? []).filter(
-    (c) => c.value !== "all",
-  );
-  return [
-    {
-      id: "default",
-      alt,
-      size,
-      contentType,
-    },
-    ...chains.map((c) => ({
-      id: c.value,
-      alt: `${alt}. ${c.label}`,
-      size,
-      contentType,
-    })),
-  ];
+// Emit a SINGLE OG image for the aggregate bench page. Previously we
+// fanned out one entry per chain variant so `/benchmarks/{slug}?chain=X`
+// social shares could render a chain-honest card, but Next hoists every
+// generateImageMetadata entry into a separate <meta property="og:image">
+// on the parent page — X, LinkedIn, Slack pick nondeterministically
+// among them, so a share of `?chain=solana` might render the `robinhood`
+// card. Chain-scoped pages live at `/benchmarks/[slug]/[chain]` and can
+// carry their own opengraph-image handler if per-chain cards become a
+// priority; today they inherit the site-wide default OG.
+export async function generateImageMetadata() {
+  return [{ id: "default", alt, size, contentType }];
 }
 
 export default async function OG({

src/app/not-found.tsx

@@ -4,7 +4,17 @@ import { ArrowLeft } from "lucide-react";
 
 export const metadata: Metadata = {
   title: "Page not found",
-  robots: { index: false },
+  // Full robots shape so Next replaces (not merges) the layout's robots
+  // meta. Previous shape `{ index: false }` caused two `<meta name="robots"
+  // content="noindex">` tags on prod 404s (layout emitted one from its
+  // own defaults, this route emitted a second) — Bing had been observed
+  // to ignore the duplicated tag and fall through to the layout's
+  // index=true. Match the shape used by the layout's IS_STAGING branch.
+  robots: {
+    index: false,
+    follow: false,
+    googleBot: { index: false, follow: false },
+  },
 };
 
 export default function NotFound() {

src/middleware.ts

@@ -69,6 +69,17 @@ const COMPARE_PATH = /^\/compare\/([a-z0-9][a-z0-9-]{0,79})\/?$/;
 export function middleware(req: NextRequest) {
   const { pathname, search } = req.nextUrl;
 
+  // Case normalisation. `/products/Alchemy` and `/Benchmarks/foo` used
+  // to serve 200 with the mixed-case URL while the HTML canonical
+  // pointed at the lowercase form — Google indexed the mixed-case URL
+  // as a valid variant and burned crawl budget on both. Force lowercase
+  // via 308 so only the canonical shape reaches origin.
+  if (pathname !== pathname.toLowerCase()) {
+    const url = req.nextUrl.clone();
+    url.pathname = pathname.toLowerCase();
+    return NextResponse.redirect(url, 308);
+  }
+
   if (search && CANONICAL_NO_QUERY.has(pathname)) {
     const canonical = req.nextUrl.clone();
     canonical.search = "";
@@ -136,6 +147,11 @@ export const config = {
     "/api/llm-context",
     "/api/freshness",
     "/api/openapi.json",
+    // Wildcard `/((?!_next|.*\..*).*)` covers every HTML route so the
+    // case-normalisation redirect fires on any URL, not just the
+    // enumerated bench/answer/compare/product paths. Excludes _next
+    // internals and any file with an extension (assets, RSS, sitemap).
+    "/((?!_next|api|.*\\..*).*)",
     "/benchmarks/:slug*",
     "/answers/:slug*",
     "/compare/:slug*",