mcp-scraper 0.7.0 → 0.8.0

package/dist/bin/api-server.cjs

@@ -6717,7 +6717,8 @@ async function spiderSite(opts) {
     totalFound: results.length,
     durationMs: Date.now() - startMs,
     truncated: queue.length > 0 || results.length >= maxUrls,
-    browserRetries
+    browserRetries,
+    sitemapUrls: sitemapUrls.slice(0, maxUrls)
   };
 }
 var SKIP_EXTENSIONS, SKIP_PATH_SEGMENTS, SKIP_PATH_PREFIXES, SKIP_QUERY_PARAMS, KERNEL_RETRY_LIMIT, UA;
@@ -6944,682 +6945,1158 @@ var init_rotating_proxy_crawl = __esm({
   }
 });
 
-// src/api/site-extractor.ts
-function estimatePixels(s) {
-  let px = 0;
-  for (const ch of s) px += PIXEL_WIDTHS[ch] ?? (ch === ch.toUpperCase() && ch !== ch.toLowerCase() ? 10 : 8);
-  return px;
-}
-function emptyPageData(url, status, via) {
-  return {
-    url,
-    status,
-    via,
-    contentType: null,
-    responseTimeMs: null,
-    sizeBytes: null,
-    redirectUrl: null,
-    lastModified: null,
-    xRobotsTag: null,
-    title: null,
-    titleLength: null,
-    titlePixels: null,
-    metaDescription: null,
-    metaDescLength: null,
-    metaKeywords: null,
-    h1: null,
-    h1_2: null,
-    h2Count: 0,
-    headings: [],
-    wordCount: 0,
-    textRatio: 0,
-    contentHash: "",
-    metaRobots: null,
-    relNext: null,
-    relPrev: null,
-    hreflang: [],
-    og: null,
-    twitter: null,
-    ampHref: null,
-    imageCount: 0,
-    imagesMissingAlt: 0,
-    imageLinks: [],
-    indexable: status === 200,
-    indexabilityReason: status === 200 ? null : "non-200",
-    schemaTypes: [],
-    canonicalUrl: null,
-    internalLinks: 0,
-    externalLinks: 0,
-    outlinks: [],
-    bodyMarkdown: "",
-    schema: []
-  };
-}
-function makeSeedSpider(startUrl, urls) {
-  return {
-    startUrl,
-    urls: urls.map((url) => ({ url, status: 200, linkedFrom: "seed", via: "fetch" })),
-    totalFound: urls.length,
-    durationMs: 0,
-    truncated: false,
-    browserRetries: 0
-  };
+// src/api/image-audit.ts
+function formatBytes(n) {
+  if (n == null) return null;
+  if (n < 1024) return `${n} B`;
+  if (n < 1024 * 1024) return `${(n / 1024).toFixed(1)} KB`;
+  return `${(n / (1024 * 1024)).toFixed(2)} MB`;
 }
-function parsePageData(url, html, status, via, resp) {
-  const origin = new URL(url).origin;
-  const headers = resp?.headers ?? {};
-  const contentType = headers["content-type"] ?? null;
-  const lastModified = headers["last-modified"] ?? null;
-  const xRobotsTag = headers["x-robots-tag"] ?? null;
-  const responseTimeMs = resp?.responseTimeMs ?? null;
-  const redirectUrl = resp?.redirectUrl ?? headers["location"] ?? null;
-  const sizeBytes = headers["content-length"] ? Number(headers["content-length"]) : Buffer.byteLength(html);
-  const title = html.match(/<title[^>]*>([^<]*)<\/title>/i)?.[1]?.trim() ?? null;
-  const metaDescription = html.match(/<meta[^>]+name=["']description["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim() ?? html.match(/<meta[^>]+content=["']([^"']*)["'][^>]+name=["']description["']/i)?.[1]?.trim() ?? null;
-  const canonicalUrl = html.match(/<link[^>]+rel=["']canonical["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? html.match(/<link[^>]+href=["']([^"']*)["'][^>]+rel=["']canonical["']/i)?.[1]?.trim() ?? null;
-  const headings = [];
-  for (const m of html.matchAll(/<h([1-6])[^>]*>([\s\S]*?)<\/h\1>/gi)) {
-    const text = m[2].replace(/<[^>]+>/g, "").trim();
-    if (text) headings.push({ level: Number(m[1]), text: text.slice(0, 200) });
-  }
-  const h1 = headings.find((h) => h.level === 1)?.text ?? null;
-  const bodyText = html.replace(/<script[\s\S]*?<\/script>/gi, "").replace(/<style[\s\S]*?<\/style>/gi, "").replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim();
-  const wordCount2 = bodyText.split(" ").filter((w) => w.length > 2).length;
-  const schemaTypes = [];
-  const schema = [];
-  for (const m of html.matchAll(/<script[^>]+type=["']application\/ld\+json["'][^>]*>([\s\S]*?)<\/script>/gi)) {
-    try {
-      const parsed = JSON.parse(m[1]);
-      if (Array.isArray(parsed)) schema.push(...parsed);
-      else schema.push(parsed);
-      const collect = (obj) => {
-        if (!obj || typeof obj !== "object") return;
-        const o = obj;
-        if (typeof o["@type"] === "string") schemaTypes.push(o["@type"]);
-        else if (Array.isArray(o["@type"])) schemaTypes.push(...o["@type"].filter((t) => typeof t === "string"));
-        if (Array.isArray(o["@graph"])) o["@graph"].forEach(collect);
-      };
-      collect(parsed);
-    } catch {
-    }
+function collectUrls(pages) {
+  const byKey = /* @__PURE__ */ new Map();
+  for (const p of pages) for (const raw of p.imageLinks || []) {
+    const u = decodeEntities(raw);
+    if (!/^https?:\/\//i.test(u)) continue;
+    if (/[{}]|%7[bd]/i.test(u)) continue;
+    const k = dedupKey(u);
+    const prev = byKey.get(k);
+    if (!prev || /^https:/i.test(u) && /^http:/i.test(prev)) byKey.set(k, u);
   }
-  const mainHtml = (html.match(/<main[^>]*>([\s\S]*?)<\/main>/i)?.[1] ?? html.match(/<article[^>]*>([\s\S]*?)<\/article>/i)?.[1] ?? html.match(/<body[^>]*>([\s\S]*?)<\/body>/i)?.[1] ?? html).replace(/<(script|style|nav|header|footer|aside|noscript|svg|iframe)[^>]*>[\s\S]*?<\/\1>/gi, "");
-  let bodyMarkdown = "";
+  return [...byKey.values()];
+}
+async function sizeAndType(url, timeoutMs) {
+  const ctrl = new AbortController();
+  const t = setTimeout(() => ctrl.abort(), timeoutMs);
+  const read = (res) => ({
+    len: res.headers.get("content-length"),
+    cr: res.headers.get("content-range"),
+    ct: res.headers.get("content-type"),
+    status: res.status
+  });
   try {
-    bodyMarkdown = turndown.turndown(mainHtml).replace(/\n{3,}/g, "\n\n").trim().slice(0, MAX_PAGE_MARKDOWN);
-  } catch {
-    bodyMarkdown = bodyText.slice(0, MAX_PAGE_MARKDOWN);
-  }
-  const outlinks = [];
-  for (const m of html.matchAll(/<a\b([^>]*?)href\s*=\s*["']([^"'#][^"']*)["']([^>]*)>([\s\S]*?)<\/a>/gi)) {
+    let bytes = null;
+    let ct = null;
+    let status = null;
     try {
-      const abs = new URL(m[2], url);
-      if (abs.protocol !== "http:" && abs.protocol !== "https:") continue;
-      const attrs = `${m[1]} ${m[3]}`;
-      const rel = attrs.match(/\brel\s*=\s*["']([^"']*)["']/i)?.[1]?.toLowerCase() ?? null;
-      const anchor = m[4].replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim().slice(0, 160);
-      outlinks.push({ href: abs.href, anchor, rel, internal: abs.origin === origin });
+      const h = read(await fetch(url, { method: "HEAD", redirect: "follow", signal: ctrl.signal }));
+      status = h.status;
+      ct = h.ct;
+      if (h.len) bytes = Number(h.len);
     } catch {
     }
-  }
-  const internalLinks = outlinks.filter((l) => l.internal).length;
-  const externalLinks = outlinks.length - internalLinks;
-  const titleLength = title?.length ?? null;
-  const titlePixels = title ? estimatePixels(title) : null;
-  const metaDescLength = metaDescription?.length ?? null;
-  const metaKeywords = html.match(/<meta[^>]+name=["']keywords["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
-  const h1List = headings.filter((h) => h.level === 1);
-  const h1_2 = h1List[1]?.text ?? null;
-  const h2Count = headings.filter((h) => h.level === 2).length;
-  const textRatio = Number((bodyText.length / Math.max(1, html.length)).toFixed(3));
-  const contentHash = (0, import_node_crypto3.createHash)("sha1").update(bodyText).digest("hex");
-  const metaRobots = html.match(/<meta[^>]+name=["']robots["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim().toLowerCase() ?? null;
-  const relNext = html.match(/<link[^>]+rel=["']next["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
-  const relPrev = html.match(/<link[^>]+rel=["']prev["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
-  const hreflang = [];
-  for (const m of html.matchAll(/<link[^>]+rel=["']alternate["'][^>]*>/gi)) {
-    const lang = m[0].match(/hreflang=["']([^"']*)["']/i)?.[1];
-    const href = m[0].match(/href=["']([^"']*)["']/i)?.[1];
-    if (lang && href) hreflang.push({ lang, href });
-  }
-  const ogTag = (p) => html.match(new RegExp(`<meta[^>]+property=["']og:${p}["'][^>]+content=["']([^"']*)["']`, "i"))?.[1]?.trim();
-  const og = ogTag("title") || ogTag("description") || ogTag("image") || ogTag("type") ? { title: ogTag("title"), description: ogTag("description"), image: ogTag("image"), type: ogTag("type") } : null;
-  const twTag = (p) => html.match(new RegExp(`<meta[^>]+name=["']twitter:${p}["'][^>]+content=["']([^"']*)["']`, "i"))?.[1]?.trim();
-  const twitter = twTag("card") || twTag("title") || twTag("description") ? { card: twTag("card"), title: twTag("title"), description: twTag("description") } : null;
-  const ampHref = html.match(/<link[^>]+rel=["']amphtml["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
-  let imageCount = 0;
-  let imagesMissingAlt = 0;
-  const imageLinkSet = /* @__PURE__ */ new Set();
-  for (const m of html.matchAll(/<img\b([^>]*)>/gi)) {
-    imageCount++;
-    const alt = m[1].match(/\balt\s*=\s*["']([^"']*)["']/i)?.[1];
-    if (!alt || !alt.trim()) imagesMissingAlt++;
-    const src = m[1].match(/\bsrc\s*=\s*["']([^"']+)["']/i)?.[1];
-    if (src) {
+    if (bytes == null) {
+      const g = await fetch(url, { method: "GET", headers: { Range: "bytes=0-0" }, redirect: "follow", signal: ctrl.signal });
+      const r = read(g);
+      status = status ?? r.status;
+      ct = ct ?? r.ct;
+      if (r.cr && r.cr.includes("/")) {
+        const tail = r.cr.split("/")[1];
+        if (tail && tail !== "*") bytes = Number(tail);
+      } else if (r.len) bytes = Number(r.len);
       try {
-        const abs = new URL(src, url);
-        if (abs.protocol === "http:" || abs.protocol === "https:") imageLinkSet.add(abs.href);
+        await g.body?.cancel();
       } catch {
       }
     }
+    return { url, status, bytes: Number.isFinite(bytes) ? bytes : null, contentType: ct };
+  } catch (e) {
+    return { url, status: null, bytes: null, contentType: null, error: String(e.message || e) };
+  } finally {
+    clearTimeout(t);
   }
-  if (og?.image) {
-    try {
-      imageLinkSet.add(new URL(og.image, url).href);
-    } catch {
+}
+async function pool(items, n, fn) {
+  const out = new Array(items.length);
+  let i = 0;
+  await Promise.all(Array.from({ length: Math.min(n, items.length) }, async () => {
+    while (i < items.length) {
+      const idx = i++;
+      out[idx] = await fn(items[idx]);
     }
-  }
-  const imageLinks = [...imageLinkSet].slice(0, 300);
-  const canonNorm = (u) => {
-    try {
-      const x = new URL(u, url);
-      return (x.origin + x.pathname).replace(/\/+$/, "");
-    } catch {
-      return u.replace(/\/+$/, "");
+  }));
+  return out;
+}
+async function auditImages(pages, opts = {}) {
+  const concurrency = opts.concurrency ?? 12;
+  const timeoutMs = opts.timeoutMs ?? 12e3;
+  const urls = collectUrls(pages).slice(0, opts.max ?? 5e3);
+  const heads = await pool(urls, concurrency, (u) => sizeAndType(u, timeoutMs));
+  const rows = heads.map((r) => {
+    const format = formatOf(r.contentType, r.url);
+    return {
+      ...r,
+      size: formatBytes(r.bytes),
+      format,
+      over100kb: r.bytes != null && r.bytes > OVER_BYTES,
+      legacyFormat: format !== "unknown" && !MODERN.has(format)
+    };
+  });
+  const sized = rows.filter((r) => r.bytes != null);
+  const totalBytes = sized.reduce((a, r) => a + r.bytes, 0);
+  const formatCounts = {};
+  for (const r of rows) formatCounts[r.format] = (formatCounts[r.format] || 0) + 1;
+  return {
+    rows: rows.sort((a, b) => (b.bytes ?? 0) - (a.bytes ?? 0)),
+    summary: {
+      unique: rows.length,
+      sized: sized.length,
+      totalBytes,
+      totalSize: formatBytes(totalBytes) ?? "0 B",
+      avgSize: formatBytes(Math.round(totalBytes / (sized.length || 1))) ?? "0 B",
+      over100kb: rows.filter((r) => r.over100kb).length,
+      legacyFormat: rows.filter((r) => r.legacyFormat).length,
+      formatCounts
     }
   };
-  const canonicalSelf = canonicalUrl ? canonNorm(canonicalUrl) === canonNorm(url) : true;
-  let indexable = true;
-  let indexabilityReason = null;
-  if (status !== 200) {
-    indexable = false;
-    indexabilityReason = "non-200";
-  } else if (metaRobots?.includes("noindex")) {
-    indexable = false;
-    indexabilityReason = "noindex";
-  } else if (xRobotsTag?.toLowerCase().includes("noindex")) {
-    indexable = false;
-    indexabilityReason = "x-robots-noindex";
-  } else if (canonicalUrl && !canonicalSelf) {
-    indexable = false;
-    indexabilityReason = "canonicalised";
-  }
-  return {
-    url,
-    status,
-    via,
-    contentType,
-    responseTimeMs,
-    sizeBytes,
-    redirectUrl,
-    lastModified,
-    xRobotsTag,
-    title,
-    titleLength,
-    titlePixels,
-    metaDescription,
-    metaDescLength,
-    metaKeywords,
-    h1,
-    h1_2,
-    h2Count,
-    headings,
-    wordCount: wordCount2,
-    textRatio,
-    contentHash,
-    metaRobots,
-    relNext,
-    relPrev,
-    hreflang,
-    og,
-    twitter,
-    ampHref,
-    imageCount,
-    imagesMissingAlt,
-    imageLinks,
-    indexable,
-    indexabilityReason,
-    schemaTypes,
-    canonicalUrl,
-    internalLinks,
-    externalLinks,
-    outlinks,
-    bodyMarkdown,
-    schema
-  };
-}
-async function extractPagesRotating(urls, opts) {
-  const fetched = await crawlWithRotation(urls, { apiKey: opts.kernelApiKey, concurrency: opts.concurrency, urlsPerBrowser: opts.urlsPerBrowser });
-  return fetched.map((r) => r.html ? parsePageData(r.url, r.html, r.status || 200, "browser", { headers: r.headers, responseTimeMs: r.responseTimeMs, redirectUrl: r.redirectUrl }) : emptyPageData(r.url, r.status || null, "browser"));
 }
-async function discoverUrls(startUrl, maxPages, kernelApiKey) {
-  const sitemapUrls = await discoverSitemapUrls2(startUrl, maxPages);
-  if (sitemapUrls.length > 0) return sitemapUrls;
-  const spider = await spiderSite({ startUrl, maxUrls: maxPages, concurrency: 12, kernelApiKey });
-  return spider.urls.filter((u) => u.status === 200 || u.status === null).slice(0, maxPages).map((u) => u.url);
+function renderImageSection(audit) {
+  const s = audit.summary;
+  const fmt = Object.entries(s.formatCounts).sort((a, b) => b[1] - a[1]).map(([k, v]) => `${k}:${v}`).join("  ");
+  const heaviest = audit.rows.filter((r) => r.over100kb).slice(0, 15);
+  const lines = [
+    `## Images`,
+    `**${s.unique} unique images** \xB7 ${s.totalSize} total \xB7 ${s.avgSize} avg \xB7 ${s.over100kb} over 100 KB \xB7 ${s.legacyFormat} legacy format`,
+    `Formats: ${fmt}`
+  ];
+  if (heaviest.length) {
+    lines.push("", "| Size | Format | URL |", "|------|--------|-----|");
+    for (const r of heaviest) lines.push(`| ${r.size} | ${r.format} | ${r.url} |`);
+  }
+  return lines.join("\n");
 }
-async function fetchAndParse(url, kernelApiKey) {
-  const startedAt = Date.now();
-  try {
-    const res = await fetch(url, {
-      headers: { "User-Agent": UA2, "Accept": "text/html,application/xhtml+xml" },
-      signal: AbortSignal.timeout(1e4),
-      redirect: "manual"
-    });
-    const meta = {
-      headers: Object.fromEntries([...res.headers.entries()].map(([k, v]) => [k.toLowerCase(), v])),
-      responseTimeMs: Date.now() - startedAt,
-      redirectUrl: res.headers.get("location")
-    };
-    if (res.ok) {
-      const ct = res.headers.get("content-type") ?? "";
-      if (ct.includes("html")) {
-        const html = await res.text();
-        return parsePageData(url, html, res.status, "fetch", meta);
-      }
-      return { ...emptyPageData(url, res.status, "fetch"), contentType: ct || null, responseTimeMs: meta.responseTimeMs ?? null };
-    }
-    if ((res.status === 403 || res.status === 429) && kernelApiKey) {
-      const html = await fetchWithKernel(url);
-      return parsePageData(url, html, 200, "browser");
-    }
-    return { ...emptyPageData(url, res.status, "fetch"), responseTimeMs: meta.responseTimeMs ?? null, redirectUrl: meta.redirectUrl ?? null };
-  } catch {
-    if (kernelApiKey) {
-      try {
-        const html = await fetchWithKernel(url);
-        return parsePageData(url, html, 200, "browser");
-      } catch {
+var OVER_BYTES, MODERN, decodeEntities, dedupKey, formatOf;
+var init_image_audit = __esm({
+  "src/api/image-audit.ts"() {
+    "use strict";
+    OVER_BYTES = 100 * 1024;
+    MODERN = /* @__PURE__ */ new Set(["webp", "avif", "svg", "svg+xml"]);
+    decodeEntities = (u) => u.replace(/&amp;/g, "&").replace(/&#0?38;/g, "&").replace(/&#x26;/gi, "&").trim();
+    dedupKey = (u) => u.replace(/^https?:\/\//i, "//").replace(/\/$/, "");
+    formatOf = (ct, url) => {
+      if (ct) {
+        const m = ct.split(";")[0].trim().toLowerCase();
+        if (m.startsWith("image/")) return m.replace("image/", "");
       }
-    }
-    return emptyPageData(url, null, "fetch");
+      return (url.split("?")[0].match(/\.([a-z0-9]+)$/i)?.[1] || "unknown").toLowerCase();
+    };
   }
+});
+
+// src/api/seo-link-graph.ts
+function normalize(u) {
+  return u.split("#")[0].replace(/\/$/, "");
 }
-async function runWithConcurrency(items, concurrency, fn) {
-  let i = 0;
-  const workers = Array.from({ length: concurrency }, async () => {
-    while (i < items.length) {
-      const item = items[i++];
-      await fn(item);
+function buildLinkGraph(pages, startUrl) {
+  const statusByUrl = /* @__PURE__ */ new Map();
+  for (const p of pages) statusByUrl.set(normalize(p.url), p.status);
+  const edges = [];
+  for (const p of pages) {
+    for (const l of p.outlinks ?? []) {
+      const nofollow = (l.rel ?? "").split(/\s+/).includes("nofollow");
+      edges.push({
+        from: p.url,
+        to: l.href,
+        anchor: l.anchor,
+        rel: l.rel,
+        internal: l.internal,
+        nofollow,
+        targetStatus: statusByUrl.get(normalize(l.href)) ?? null
+      });
     }
-  });
-  await Promise.all(workers);
-}
-async function extractSite(opts) {
-  const startMs = Date.now();
-  const maxPages = Math.min(opts.maxPages ?? 100, 1e4);
-  const concurrency = Math.min(opts.concurrency ?? EXTRACT_CONCURRENCY, 10);
-  const rotating = Boolean(opts.rotateProxyEvery && opts.kernelApiKey);
-  const seeded = opts.seedUrls && opts.seedUrls.length > 0;
-  const pages = [];
-  let browserRetries = 0;
-  let truncated = false;
-  let spider;
-  if (rotating) {
-    const browserConcurrency = Math.max(1, Math.min(opts.parallelism ?? 1, 8));
-    const urlsPerBrowser = opts.rotateProxyEvery ?? 10;
-    const followLinks = !seeded;
-    const seen = /* @__PURE__ */ new Set();
-    const frontier = [];
-    const enqueue = (u) => {
-      const n = normalizeUrl(u, opts.startUrl);
-      if (n && !seen.has(n)) {
-        seen.add(n);
-        frontier.push(n);
-      }
-    };
-    if (seeded) {
-      for (const u of opts.seedUrls) enqueue(u);
-    } else {
-      enqueue(opts.startUrl);
-      for (const u of await discoverSitemapUrls2(opts.startUrl, maxPages)) enqueue(u);
+  }
+  const inboundFrom = /* @__PURE__ */ new Map();
+  const inboundAnchors = /* @__PURE__ */ new Map();
+  const adjacency = /* @__PURE__ */ new Map();
+  for (const e of edges) {
+    if (!e.internal) continue;
+    const to = normalize(e.to);
+    const from = normalize(e.from);
+    if (!inboundFrom.has(to)) inboundFrom.set(to, /* @__PURE__ */ new Set());
+    inboundFrom.get(to).add(from);
+    if (e.anchor) {
+      if (!inboundAnchors.has(to)) inboundAnchors.set(to, []);
+      inboundAnchors.get(to).push(e.anchor);
     }
-    const waveCap = Math.max(browserConcurrency * urlsPerBrowser, 30);
-    let done = 0;
-    while (frontier.length > 0 && pages.length < maxPages) {
-      const capacity = maxPages - pages.length;
-      const wave = frontier.splice(0, Math.min(capacity, waveCap));
-      const fetched = await crawlWithRotation(wave, {
-        apiKey: opts.kernelApiKey,
-        concurrency: browserConcurrency,
-        urlsPerBrowser,
-        onPage: (p) => {
-          done++;
-          opts.onProgress?.({ done, discovered: seen.size, frontier: frontier.length, lastOk: p.ok ? 1 : 0, lastFail: p.ok ? 0 : 1, phase: p.phase });
-        }
-      });
-      for (const r of fetched) {
-        const page = r.html ? parsePageData(r.url, r.html, r.status || 200, "browser", { headers: r.headers, responseTimeMs: r.responseTimeMs, redirectUrl: r.redirectUrl }) : emptyPageData(r.url, r.status || null, "browser");
-        pages.push(page);
-        browserRetries++;
-        if (followLinks && r.html) {
-          for (const l of page.outlinks) if (l.internal) enqueue(l.href);
-        }
-      }
+    if (!e.nofollow && (statusByUrl.get(to) === 200 || !statusByUrl.has(to))) {
+      if (!adjacency.has(from)) adjacency.set(from, /* @__PURE__ */ new Set());
+      adjacency.get(from).add(to);
     }
-    truncated = frontier.length > 0;
-    spider = makeSeedSpider(opts.startUrl, [...seen]);
-  } else {
-    spider = seeded ? makeSeedSpider(opts.startUrl, opts.seedUrls) : await spiderSite({ startUrl: opts.startUrl, maxUrls: maxPages, concurrency: 12, kernelApiKey: opts.kernelApiKey });
-    const urlsToExtract = spider.urls.filter((u) => u.status === 200 || u.status === null).slice(0, maxPages).map((u) => u.url);
-    browserRetries = spider.browserRetries;
-    await runWithConcurrency(urlsToExtract, concurrency, async (url) => {
-      const data = await fetchAndParse(url, opts.kernelApiKey);
-      if (data.via === "browser") browserRetries++;
-      pages.push(data);
-    });
-    truncated = spider.truncated || pages.length >= maxPages;
   }
-  let branding = null;
-  if (opts.formats?.includes("branding") && opts.kernelApiKey) {
-    branding = await extractBranding(opts.startUrl, opts.kernelApiKey).catch(() => null);
+  const depth = /* @__PURE__ */ new Map();
+  const start = normalize(startUrl);
+  const queue = [start];
+  depth.set(start, 0);
+  while (queue.length > 0) {
+    const cur = queue.shift();
+    const d = depth.get(cur);
+    for (const next of adjacency.get(cur) ?? []) {
+      if (!depth.has(next)) {
+        depth.set(next, d + 1);
+        queue.push(next);
+      }
+    }
   }
-  return {
-    startUrl: opts.startUrl,
-    pages,
-    totalPages: pages.length,
-    spider,
-    durationMs: Date.now() - startMs,
-    truncated,
-    browserRetries,
-    branding
+  const topAnchorsFor = (url) => {
+    const counts = /* @__PURE__ */ new Map();
+    for (const a of inboundAnchors.get(url) ?? []) counts.set(a, (counts.get(a) ?? 0) + 1);
+    return [...counts.entries()].sort((a, b) => b[1] - a[1]).slice(0, 3).map((e) => e[0]);
   };
-}
-var import_turndown2, import_node_crypto3, PIXEL_WIDTHS, UA2, EXTRACT_CONCURRENCY, MAX_PAGE_MARKDOWN, turndown;
-var init_site_extractor = __esm({
-  "src/api/site-extractor.ts"() {
-    "use strict";
-    import_turndown2 = __toESM(require("turndown"), 1);
-    import_node_crypto3 = require("crypto");
-    init_site_mapper();
-    init_kernel_fetch();
-    init_rotating_proxy_crawl();
-    init_screenshot();
-    PIXEL_WIDTHS = { i: 4, l: 4, j: 4, ".": 4, ",": 4, "'": 4, t: 6, f: 6, r: 6, " ": 4, m: 14, w: 13, W: 16, M: 16 };
-    UA2 = "Mozilla/5.0 (compatible; ThorbitBot/1.0; +https://thorbit.ai)";
-    EXTRACT_CONCURRENCY = 6;
-    MAX_PAGE_MARKDOWN = 4e4;
-    turndown = new import_turndown2.default({ headingStyle: "atx", bulletListMarker: "-", codeBlockStyle: "fenced" });
+  const metrics = /* @__PURE__ */ new Map();
+  for (const p of pages) {
+    const key = normalize(p.url);
+    const inbound = inboundFrom.get(key) ?? /* @__PURE__ */ new Set();
+    metrics.set(p.url, {
+      url: p.url,
+      inlinks: inbound.size,
+      uniqueInlinks: inbound.size,
+      outlinksInternal: (p.outlinks ?? []).filter((l) => l.internal).length,
+      outlinksExternal: (p.outlinks ?? []).filter((l) => !l.internal).length,
+      crawlDepth: depth.has(key) ? depth.get(key) : null,
+      orphan: inbound.size === 0 && key !== start,
+      topAnchors: topAnchorsFor(key)
+    });
   }
-});
-
-// src/inngest/client.ts
-var import_inngest, inngest;
-var init_client = __esm({
-  "src/inngest/client.ts"() {
+  return { edges, metrics };
+}
+var init_seo_link_graph = __esm({
+  "src/api/seo-link-graph.ts"() {
     "use strict";
-    import_inngest = require("inngest");
-    inngest = new import_inngest.Inngest({ id: "mcp-scraper" });
   }
 });
 
-// src/lib/llm-parse-with-retry.ts
-function extractJson(raw) {
-  let text = raw.replace(/<think>[\s\S]*?<\/think>/gi, "").trim();
-  text = text.replace(/^```(?:json)?\s*/i, "").replace(/\s*```\s*$/, "").trim();
-  if (text.startsWith("{") || text.startsWith("[")) return text;
-  const objStart = text.indexOf("{");
-  const arrStart = text.indexOf("[");
-  let start = -1;
-  let open, close;
-  if (objStart === -1 && arrStart === -1) return text;
-  if (objStart === -1) {
-    start = arrStart;
-    open = "[";
-    close = "]";
-  } else if (arrStart === -1) {
-    start = objStart;
-    open = "{";
-    close = "}";
-  } else if (arrStart < objStart) {
-    start = arrStart;
-    open = "[";
-    close = "]";
-  } else {
-    start = objStart;
-    open = "{";
-    close = "}";
+// src/api/seo-link-report.ts
+function registrableDomain(host) {
+  const h = host.replace(/^www\./, "").toLowerCase();
+  const parts = h.split(".");
+  return parts.length <= 2 ? h : parts.slice(-2).join(".");
+}
+function buildLinkReport(edges, metrics, siteUrl) {
+  let siteReg = "";
+  try {
+    siteReg = registrableDomain(new URL(siteUrl).hostname);
+  } catch {
+    siteReg = "";
   }
-  let depth = 0;
-  let inStr = false;
-  let escape = false;
-  for (let i = start; i < text.length; i++) {
-    const ch = text[i];
-    if (escape) {
-      escape = false;
-      continue;
-    }
-    if (ch === "\\" && inStr) {
-      escape = true;
-      continue;
-    }
-    if (ch === '"') {
-      inStr = !inStr;
-      continue;
-    }
-    if (inStr) continue;
-    if (ch === open) depth++;
-    else if (ch === close) {
-      if (--depth === 0) return text.slice(start, i + 1);
-    }
+  const internalEdges = edges.filter((e) => e.internal);
+  const pages = metrics.length;
+  const orphans = metrics.filter((m) => m.orphan).length;
+  const brokenInternal = internalEdges.filter((e) => e.targetStatus != null && e.targetStatus >= 400).length;
+  const sumInlinks = metrics.reduce((a, m) => a + m.inlinks, 0);
+  const sumOutlinks = metrics.reduce((a, m) => a + m.outlinksInternal + m.outlinksExternal, 0);
+  const distribution = { zero: 0, oneToTwo: 0, threeToTen: 0, elevenPlus: 0 };
+  for (const m of metrics) {
+    if (m.inlinks === 0) distribution.zero++;
+    else if (m.inlinks <= 2) distribution.oneToTwo++;
+    else if (m.inlinks <= 10) distribution.threeToTen++;
+    else distribution.elevenPlus++;
   }
-  return text.slice(start);
-}
-async function llmParseWithRetry(schema, promptFn, llm) {
-  let lastError;
-  for (let attempt = 0; attempt < 3; attempt++) {
-    const prompt = attempt === 0 ? promptFn(attempt) : promptFn(attempt, lastError?.message);
-    let raw;
+  const topByInlinks = [...metrics].sort((a, b) => b.inlinks - a.inlinks).slice(0, 20).map((m) => ({ url: m.url, inlinks: m.inlinks, outlinksInternal: m.outlinksInternal, outlinksExternal: m.outlinksExternal }));
+  const domMap = /* @__PURE__ */ new Map();
+  let externalTotal = 0;
+  for (const e of edges) {
+    let domain;
     try {
-      raw = await llm.completeJson(prompt);
-    } catch (e) {
-      lastError = e instanceof Error ? e : new Error(String(e));
+      domain = registrableDomain(new URL(e.to).hostname);
+    } catch {
       continue;
     }
-    const result = schema.safeParse(raw);
-    if (result.success) return result.data;
-    lastError = result.error;
+    if (!domain || domain === siteReg) continue;
+    externalTotal++;
+    const d = domMap.get(domain) ?? { links: 0, nofollow: 0, pages: /* @__PURE__ */ new Set() };
+    d.links++;
+    if (e.nofollow) d.nofollow++;
+    d.pages.add(e.from);
+    domMap.set(domain, d);
   }
-  const detail = lastError?.message?.slice(0, 200) ?? "unknown error";
-  throw new Error(`LLM failed to produce valid output after 3 attempts: ${detail}`);
+  const externalDomains = [...domMap.entries()].map(([domain, d]) => ({ domain, links: d.links, nofollow: d.nofollow, pages: d.pages.size })).sort((a, b) => b.links - a.links);
+  const round = (n) => Math.round(n * 10) / 10;
+  return {
+    summary: {
+      internal: {
+        totalLinks: internalEdges.length,
+        pages,
+        orphans,
+        brokenInternal,
+        avgInlinks: pages ? round(sumInlinks / pages) : 0,
+        avgOutlinks: pages ? round(sumOutlinks / pages) : 0,
+        distribution,
+        topByInlinks
+      },
+      external: {
+        totalLinks: externalTotal,
+        uniqueDomains: externalDomains.length,
+        topDomains: externalDomains.slice(0, 20)
+      }
+    },
+    externalDomains
+  };
 }
-var import_sdk4, DeepInfraLlmClient, OpenRouterLlmClient, FallbackLlmClient;
-var init_llm_parse_with_retry = __esm({
-  "src/lib/llm-parse-with-retry.ts"() {
+function renderLinkReport(r) {
+  const s = r.summary;
+  const lines = [
+    `## Link analysis`,
+    `**Internal:** ${s.internal.totalLinks} links \xB7 ${s.internal.pages} pages \xB7 avg ${s.internal.avgInlinks} inlinks/page \xB7 ${s.internal.orphans} orphans \xB7 ${s.internal.brokenInternal} broken`,
+    `**External:** ${s.external.totalLinks} links to ${s.external.uniqueDomains} domains`,
+    ``,
+    `### Inlink distribution`,
+    `- 0 inlinks (orphans): ${s.internal.distribution.zero}`,
+    `- 1\u20132: ${s.internal.distribution.oneToTwo}`,
+    `- 3\u201310: ${s.internal.distribution.threeToTen}`,
+    `- 11+: ${s.internal.distribution.elevenPlus}`,
+    ``,
+    `### Top 20 internal pages by inlinks`,
+    `| inlinks | out (int/ext) | URL |`,
+    `|---|---|---|`,
+    ...s.internal.topByInlinks.map((p) => `| ${p.inlinks} | ${p.outlinksInternal}/${p.outlinksExternal} | ${p.url} |`),
+    ``,
+    `### Top 20 external domains by links`,
+    `| links | nofollow | from pages | domain |`,
+    `|---|---|---|---|`,
+    ...s.external.topDomains.map((d) => `| ${d.links} | ${d.nofollow} | ${d.pages} | ${d.domain} |`)
+  ];
+  return lines.join("\n");
+}
+var init_seo_link_report = __esm({
+  "src/api/seo-link-report.ts"() {
     "use strict";
-    import_sdk4 = __toESM(require("@anthropic-ai/sdk"), 1);
-    init_cost_telemetry();
-    DeepInfraLlmClient = class {
-      apiKey;
-      model;
-      totalCostUsd = 0;
-      constructor(apiKey, model = "Qwen/Qwen3.6-35B-A3B") {
-        this.apiKey = apiKey;
-        this.model = model;
-      }
-      async complete(prompt) {
-        const resp = await fetch("https://api.deepinfra.com/v1/openai/chat/completions", {
-          method: "POST",
-          headers: { "Content-Type": "application/json", "Authorization": `Bearer ${this.apiKey}` },
-          body: JSON.stringify({ model: this.model, messages: [{ role: "user", content: prompt }], max_tokens: 8192 })
-        });
-        if (!resp.ok) throw new Error(`DeepInfra error ${resp.status}: ${await resp.text()}`);
-        const data = await resp.json();
-        void recordVendorUsage({ vendor: "deepinfra_qwen", model: this.model, units: data.usage?.completion_tokens ?? 0, unitType: "tokens" });
-        return data.choices[0].message.content;
-      }
-      async completeJson(prompt) {
-        const raw = await this.complete(prompt);
-        try {
-          return JSON.parse(extractJson(raw));
-        } catch {
-          const coerced = await this.complete(
-            `Convert the following text to a valid JSON object. Output ONLY the JSON, no markdown, no explanation:
+  }
+});
 
-${raw}`
-          );
-          return JSON.parse(extractJson(coerced));
-        }
-      }
-    };
-    OpenRouterLlmClient = class {
-      apiKey;
-      model;
-      totalCostUsd = 0;
-      constructor(apiKey, model = "qwen/qwen3-235b-a22b") {
-        this.apiKey = apiKey;
-        this.model = model;
-      }
-      async complete(prompt) {
-        const resp = await fetch("https://openrouter.ai/api/v1/chat/completions", {
-          method: "POST",
-          headers: { "Content-Type": "application/json", "Authorization": `Bearer ${this.apiKey}` },
-          body: JSON.stringify({ model: this.model, messages: [{ role: "user", content: prompt }], max_tokens: 8192 })
-        });
-        if (!resp.ok) throw new Error(`OpenRouter error ${resp.status}: ${await resp.text()}`);
-        const data = await resp.json();
-        const cost = data.usage?.cost ?? 0;
-        this.totalCostUsd += cost;
-        void recordVendorUsage({ vendor: "openrouter", model: this.model, units: cost, unitType: "usd" });
-        return data.choices[0].message.content;
-      }
-      async completeJson(prompt) {
-        const raw = await this.complete(prompt);
-        try {
-          return JSON.parse(extractJson(raw));
-        } catch {
-          const coerced = await this.complete(
-            `Convert the following text to a valid JSON object. Output ONLY the JSON, no markdown, no explanation:
-
-${raw}`
-          );
-          return JSON.parse(extractJson(coerced));
-        }
+// src/api/seo-issues.ts
+function dupes(pages, key) {
+  const groups = /* @__PURE__ */ new Map();
+  for (const p of pages) {
+    const k = key(p);
+    if (k === null || k === void 0 || k === "") continue;
+    const ks = String(k);
+    if (!groups.has(ks)) groups.set(ks, []);
+    groups.get(ks).push(p.url);
+  }
+  return [...groups.values()].filter((g) => g.length > 1).flat();
+}
+function computeIssues(pages, metrics, precomputedBrokenLinkPages) {
+  const group = (urls) => ({ count: urls.length, urls: urls.slice(0, 200) });
+  const where = (fn) => pages.filter(fn).map((p) => p.url);
+  const ok = (p) => p.status === 200;
+  const pathname = (u) => {
+    try {
+      return new URL(u).pathname;
+    } catch {
+      return "";
+    }
+  };
+  const report = {
+    "title.missing": group(where((p) => ok(p) && !p.title)),
+    "title.duplicate": group(dupes(pages.filter(ok), (p) => p.title)),
+    "title.tooLong": group(where((p) => (p.titleLength ?? 0) > TITLE_MAX || (p.titlePixels ?? 0) > TITLE_PX_MAX)),
+    "title.tooShort": group(where((p) => p.title != null && (p.titleLength ?? 0) < TITLE_MIN)),
+    "title.sameAsH1": group(where((p) => !!p.title && !!p.h1 && p.title.trim() === p.h1.trim())),
+    "meta.missing": group(where((p) => ok(p) && !p.metaDescription)),
+    "meta.duplicate": group(dupes(pages.filter(ok), (p) => p.metaDescription)),
+    "meta.tooLong": group(where((p) => (p.metaDescLength ?? 0) > META_MAX)),
+    "meta.tooShort": group(where((p) => p.metaDescription != null && (p.metaDescLength ?? 0) < META_MIN)),
+    "h1.missing": group(where((p) => ok(p) && !p.h1)),
+    "h1.multiple": group(where((p) => !!p.h1_2)),
+    "h1.duplicate": group(dupes(pages.filter(ok), (p) => p.h1)),
+    "h1.tooLong": group(where((p) => (p.h1?.length ?? 0) > H1_MAX)),
+    "h2.missing": group(where((p) => ok(p) && p.h2Count === 0)),
+    "indexability.nonIndexable": group(where((p) => !p.indexable)),
+    "indexability.noindex": group(where((p) => p.indexabilityReason === "noindex")),
+    "canonical.missing": group(where((p) => ok(p) && !p.canonicalUrl)),
+    "canonical.canonicalised": group(where((p) => p.indexabilityReason === "canonicalised")),
+    "response.broken4xx": group(where((p) => (p.status ?? 0) >= 400 && (p.status ?? 0) < 500)),
+    "response.error5xx": group(where((p) => (p.status ?? 0) >= 500)),
+    "response.redirect3xx": group(where((p) => (p.status ?? 0) >= 300 && (p.status ?? 0) < 400)),
+    "response.noResponse": group(where((p) => p.status === null)),
+    "content.thin": group(where((p) => ok(p) && p.wordCount > 0 && p.wordCount < THIN_WORDS)),
+    "content.exactDuplicate": group(dupes(pages.filter((p) => ok(p) && p.contentHash), (p) => p.contentHash)),
+    "images.missingAlt": group(where((p) => p.imagesMissingAlt > 0)),
+    "schema.missing": group(where((p) => ok(p) && p.schemaTypes.length === 0)),
+    "url.tooLong": group(where((p) => p.url.length > URL_MAX)),
+    "url.uppercase": group(where((p) => /[A-Z]/.test(pathname(p.url)))),
+    "url.underscores": group(where((p) => pathname(p.url).includes("_"))),
+    "links.orphan": group([...metrics.values()].filter((m) => m.orphan).map((m) => m.url))
+  };
+  let brokenLinkPages = precomputedBrokenLinkPages;
+  if (!brokenLinkPages) {
+    const normUrl2 = (u) => {
+      try {
+        const x = new URL(u);
+        x.hash = "";
+        return (x.origin + x.pathname).replace(/\/+$/, "") + x.search;
+      } catch {
+        return u.replace(/#.*$/, "").replace(/\/+$/, "");
       }
     };
-    FallbackLlmClient = class {
-      constructor(primary, secondary) {
-        this.primary = primary;
-        this.secondary = secondary;
-      }
-      primary;
-      secondary;
-      lastServedBy = null;
-      lastPrimaryError = null;
-      get totalCostUsd() {
-        return this.primary.totalCostUsd + this.secondary.totalCostUsd;
-      }
-      async complete(prompt) {
-        try {
-          const result = await this.primary.complete(prompt);
-          this.lastServedBy = "primary";
-          this.lastPrimaryError = null;
-          return result;
-        } catch (err) {
-          this.lastPrimaryError = String(err);
-          console.warn("[FallbackLlmClient] primary failed, using secondary:", err);
-          this.lastServedBy = "secondary";
-          return this.secondary.complete(prompt);
-        }
-      }
-      async completeJson(prompt) {
-        try {
-          const result = await this.primary.completeJson(prompt);
-          this.lastServedBy = "primary";
-          this.lastPrimaryError = null;
-          return result;
-        } catch (err) {
-          this.lastPrimaryError = String(err);
-          console.warn("[FallbackLlmClient] primary failed, using secondary:", err);
-          this.lastServedBy = "secondary";
-          return this.secondary.completeJson(prompt);
-        }
+    const statusByUrl = new Map(pages.map((p) => [normUrl2(p.url), p.status]));
+    brokenLinkPages = /* @__PURE__ */ new Set();
+    for (const p of pages) {
+      for (const l of p.outlinks ?? []) {
+        if (!l.internal) continue;
+        const st = statusByUrl.get(normUrl2(l.href));
+        if (st != null && st >= 400) brokenLinkPages.add(p.url);
       }
-    };
+    }
+  }
+  report["links.brokenInternal"] = group([...brokenLinkPages]);
+  return report;
+}
+function renderIssueReport(siteUrl, pages, report, metrics) {
+  const total = pages.length;
+  const sev = (key) => key.startsWith("response.broken") || key.startsWith("response.error") || key.startsWith("links.broken") ? "\u{1F534}" : key.startsWith("title.missing") || key.startsWith("h1.missing") || key.startsWith("indexability") || key.startsWith("canonical.missing") ? "\u{1F7E0}" : "\u{1F7E1}";
+  const rows = Object.entries(report).filter(([, g]) => g.count > 0).sort((a, b) => b[1].count - a[1].count).map(([k, g]) => `| ${sev(k)} | \`${k}\` | ${g.count} | ${g.urls.slice(0, 3).join(" \xB7 ")}${g.urls.length > 3 ? " \u2026" : ""} |`);
+  const depths = [...metrics.values()].map((m) => m.crawlDepth).filter((d) => d != null);
+  const maxDepth = depths.length ? Math.max(...depths) : 0;
+  const orphans = [...metrics.values()].filter((m) => m.orphan).length;
+  const topLinked = [...metrics.values()].sort((a, b) => b.inlinks - a.inlinks).slice(0, 5);
+  return [
+    `# SEO Crawl Report: ${siteUrl}`,
+    `**${total} pages crawled** \xB7 max crawl depth ${maxDepth} \xB7 ${orphans} orphan page(s)`,
+    `
+## Issues
+| | Issue | Count | Examples |
+|---|-------|-------|----------|
+${rows.join("\n") || "| \u2705 | none | 0 | \u2014 |"}`,
+    `
+## Most-linked pages
+${topLinked.map((m) => `- ${m.inlinks} inlinks \xB7 depth ${m.crawlDepth ?? "\u2014"} \xB7 ${m.url}`).join("\n")}`,
+    `
+_Thresholds: title ${TITLE_MAX}ch/${TITLE_PX_MAX}px, meta ${META_MAX}ch, H1 ${H1_MAX}ch, thin <${THIN_WORDS} words, URL ${URL_MAX}ch. Pixel widths are estimates._`
+  ].join("\n");
+}
+var THIN_WORDS, TITLE_MAX, TITLE_PX_MAX, TITLE_MIN, META_MAX, META_MIN, H1_MAX, URL_MAX;
+var init_seo_issues = __esm({
+  "src/api/seo-issues.ts"() {
+    "use strict";
+    THIN_WORDS = 200;
+    TITLE_MAX = 60;
+    TITLE_PX_MAX = 561;
+    TITLE_MIN = 30;
+    META_MAX = 155;
+    META_MIN = 70;
+    H1_MAX = 70;
+    URL_MAX = 115;
   }
 });
 
-// src/lib/http-client.ts
-var FetchHttpClient;
-var init_http_client = __esm({
-  "src/lib/http-client.ts"() {
+// src/api/seo-audit.ts
+function buildSeoAudit(pages, startUrl) {
+  const { edges, metrics } = buildLinkGraph(pages, startUrl);
+  const issues = computeIssues(pages, metrics);
+  const { summary } = buildLinkReport(edges, [...metrics.values()], startUrl);
+  return {
+    issues,
+    linkSummary: summary,
+    pageMetrics: [...metrics.values()].map((m) => ({
+      url: m.url,
+      inboundInternal: m.inlinks,
+      outboundInternal: m.outlinksInternal,
+      orphan: m.orphan
+    }))
+  };
+}
+var init_seo_audit = __esm({
+  "src/api/seo-audit.ts"() {
     "use strict";
-    FetchHttpClient = class {
-      async get(url, headers) {
-        const response = await fetch(url, { headers });
-        const body = await response.text();
-        return { status: response.status, body };
-      }
-      async post(url, body, headers) {
-        const response = await fetch(url, {
-          method: "POST",
-          headers: { "Content-Type": "application/json", ...headers },
-          body: JSON.stringify(body)
-        });
-        const responseBody = await response.text();
-        return { status: response.status, body: responseBody };
-      }
-    };
+    init_seo_link_graph();
+    init_seo_link_report();
+    init_seo_issues();
   }
 });
 
-// src/lib/site-architecture-auditor/rules.ts
-function applyGuardrailRules(raw, context) {
-  const passed = [];
-  const flagged = [];
-  for (const item of raw) {
-    const flags = [];
-    const rec = item;
-    const anchor = typeof rec["anchor_text"] === "string" ? rec["anchor_text"].trim().toLowerCase() : null;
-    if (anchor !== null) {
-      const isVague = VAGUE_ANCHOR_BLOCKLIST.some(
-        (blocked) => blocked.toLowerCase() === anchor
-      );
-      if (isVague) flags.push("vague_anchor");
-    }
-    const outlinks = typeof rec["outlink_count"] === "number" ? rec["outlink_count"] : null;
-    if (outlinks !== null && outlinks >= EXCESSIVE_OUTLINKS_THRESHOLD) {
-      flags.push("excessive_outlinks");
-    }
-    if (context !== null && context !== void 0 && typeof context["resolveFlags"] === "function") {
-      const resolveFlags = context["resolveFlags"];
-      const contextualCandidates = [
-        "seo_only_link",
-        "exact_match_repetition",
-        "orphaned_support_page",
-        "crawl_depth_regression",
-        "low_content_source",
-        "circular_authority_trap"
-      ];
-      const contextFlags = resolveFlags(item, contextualCandidates);
-      flags.push(...contextFlags);
-    }
-    if (flags.length === 0) {
-      passed.push(item);
-    } else {
-      flagged.push({ item, flags });
-    }
-  }
-  return { passed, flagged };
+// src/api/site-extractor.ts
+function estimatePixels(s) {
+  let px = 0;
+  for (const ch of s) px += PIXEL_WIDTHS[ch] ?? (ch === ch.toUpperCase() && ch !== ch.toLowerCase() ? 10 : 8);
+  return px;
 }
-var import_zod2, PageTypeSchema, ClassificationConfidenceSchema, ARCHITECTURE_SCORE_WEIGHTS, ILES_WEIGHTS, PriorityTierSchema, EXCESSIVE_OUTLINKS_THRESHOLD, VAGUE_ANCHOR_BLOCKLIST, GuardrailFlagNameSchema, RequiredReportSectionSchema, LocationArchitecturePatternSchema;
-var init_rules = __esm({
-  "src/lib/site-architecture-auditor/rules.ts"() {
-    "use strict";
-    import_zod2 = require("zod");
-    PageTypeSchema = import_zod2.z.enum([
-      "homepage",
-      "core_service",
-      "sub_service",
-      "location",
-      "service_location",
-      "industry",
-      "product",
-      "pricing",
-      "faq",
-      "blog_post",
-      "guide_resource",
-      "case_study",
-      "comparison",
-      "competitor",
-      "about",
-      "contact",
-      "team",
-      "careers",
-      "privacy_policy",
-      "terms",
-      "html_sitemap",
-      "login_portal",
+function emptyPageData(url, status, via) {
+  return {
+    url,
+    status,
+    via,
+    contentType: null,
+    responseTimeMs: null,
+    sizeBytes: null,
+    redirectUrl: null,
+    lastModified: null,
+    xRobotsTag: null,
+    title: null,
+    titleLength: null,
+    titlePixels: null,
+    metaDescription: null,
+    metaDescLength: null,
+    metaKeywords: null,
+    h1: null,
+    h1_2: null,
+    h2Count: 0,
+    headings: [],
+    wordCount: 0,
+    textRatio: 0,
+    contentHash: "",
+    metaRobots: null,
+    relNext: null,
+    relPrev: null,
+    hreflang: [],
+    og: null,
+    twitter: null,
+    ampHref: null,
+    imageCount: 0,
+    imagesMissingAlt: 0,
+    imageLinks: [],
+    indexable: status === 200,
+    indexabilityReason: status === 200 ? null : "non-200",
+    inSitemap: null,
+    schemaTypes: [],
+    canonicalUrl: null,
+    internalLinks: 0,
+    externalLinks: 0,
+    outlinks: [],
+    bodyMarkdown: "",
+    schema: []
+  };
+}
+function makeSeedSpider(startUrl, urls) {
+  return {
+    startUrl,
+    urls: urls.map((url) => ({ url, status: 200, linkedFrom: "seed", via: "fetch" })),
+    totalFound: urls.length,
+    durationMs: 0,
+    truncated: false,
+    browserRetries: 0,
+    sitemapUrls: []
+  };
+}
+function parsePageData(url, html, status, via, resp) {
+  const origin = new URL(url).origin;
+  const headers = resp?.headers ?? {};
+  const contentType = headers["content-type"] ?? null;
+  const lastModified = headers["last-modified"] ?? null;
+  const xRobotsTag = headers["x-robots-tag"] ?? null;
+  const responseTimeMs = resp?.responseTimeMs ?? null;
+  const redirectUrl = resp?.redirectUrl ?? headers["location"] ?? null;
+  const sizeBytes = headers["content-length"] ? Number(headers["content-length"]) : Buffer.byteLength(html);
+  const title = html.match(/<title[^>]*>([^<]*)<\/title>/i)?.[1]?.trim() ?? null;
+  const metaDescription = html.match(/<meta[^>]+name=["']description["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim() ?? html.match(/<meta[^>]+content=["']([^"']*)["'][^>]+name=["']description["']/i)?.[1]?.trim() ?? null;
+  const canonicalUrl = html.match(/<link[^>]+rel=["']canonical["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? html.match(/<link[^>]+href=["']([^"']*)["'][^>]+rel=["']canonical["']/i)?.[1]?.trim() ?? null;
+  const headings = [];
+  for (const m of html.matchAll(/<h([1-6])[^>]*>([\s\S]*?)<\/h\1>/gi)) {
+    const text = m[2].replace(/<[^>]+>/g, "").trim();
+    if (text) headings.push({ level: Number(m[1]), text: text.slice(0, 200) });
+  }
+  const h1 = headings.find((h) => h.level === 1)?.text ?? null;
+  const bodyText = html.replace(/<script[\s\S]*?<\/script>/gi, "").replace(/<style[\s\S]*?<\/style>/gi, "").replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim();
+  const wordCount2 = bodyText.split(" ").filter((w) => w.length > 2).length;
+  const schemaTypes = [];
+  const schema = [];
+  for (const m of html.matchAll(/<script[^>]+type=["']application\/ld\+json["'][^>]*>([\s\S]*?)<\/script>/gi)) {
+    try {
+      const parsed = JSON.parse(m[1]);
+      if (Array.isArray(parsed)) schema.push(...parsed);
+      else schema.push(parsed);
+      const collect = (obj) => {
+        if (!obj || typeof obj !== "object") return;
+        const o = obj;
+        if (typeof o["@type"] === "string") schemaTypes.push(o["@type"]);
+        else if (Array.isArray(o["@type"])) schemaTypes.push(...o["@type"].filter((t) => typeof t === "string"));
+        if (Array.isArray(o["@graph"])) o["@graph"].forEach(collect);
+      };
+      collect(parsed);
+    } catch {
+    }
+  }
+  const mainHtml = (html.match(/<main[^>]*>([\s\S]*?)<\/main>/i)?.[1] ?? html.match(/<article[^>]*>([\s\S]*?)<\/article>/i)?.[1] ?? html.match(/<body[^>]*>([\s\S]*?)<\/body>/i)?.[1] ?? html).replace(/<(script|style|nav|header|footer|aside|noscript|svg|iframe)[^>]*>[\s\S]*?<\/\1>/gi, "");
+  let bodyMarkdown = "";
+  try {
+    bodyMarkdown = turndown.turndown(mainHtml).replace(/\n{3,}/g, "\n\n").trim().slice(0, MAX_PAGE_MARKDOWN);
+  } catch {
+    bodyMarkdown = bodyText.slice(0, MAX_PAGE_MARKDOWN);
+  }
+  const outlinks = [];
+  for (const m of html.matchAll(/<a\b([^>]*?)href\s*=\s*["']([^"'#][^"']*)["']([^>]*)>([\s\S]*?)<\/a>/gi)) {
+    try {
+      const abs = new URL(m[2], url);
+      if (abs.protocol !== "http:" && abs.protocol !== "https:") continue;
+      const attrs = `${m[1]} ${m[3]}`;
+      const rel = attrs.match(/\brel\s*=\s*["']([^"']*)["']/i)?.[1]?.toLowerCase() ?? null;
+      const anchor = m[4].replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim().slice(0, 160);
+      outlinks.push({ href: abs.href, anchor, rel, internal: abs.origin === origin });
+    } catch {
+    }
+  }
+  const internalLinks = outlinks.filter((l) => l.internal).length;
+  const externalLinks = outlinks.length - internalLinks;
+  const titleLength = title?.length ?? null;
+  const titlePixels = title ? estimatePixels(title) : null;
+  const metaDescLength = metaDescription?.length ?? null;
+  const metaKeywords = html.match(/<meta[^>]+name=["']keywords["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
+  const h1List = headings.filter((h) => h.level === 1);
+  const h1_2 = h1List[1]?.text ?? null;
+  const h2Count = headings.filter((h) => h.level === 2).length;
+  const textRatio = Number((bodyText.length / Math.max(1, html.length)).toFixed(3));
+  const contentHash = (0, import_node_crypto3.createHash)("sha1").update(bodyText).digest("hex");
+  const metaRobots = html.match(/<meta[^>]+name=["']robots["'][^>]+content=["']([^"']*)["']/i)?.[1]?.trim().toLowerCase() ?? null;
+  const relNext = html.match(/<link[^>]+rel=["']next["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
+  const relPrev = html.match(/<link[^>]+rel=["']prev["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
+  const hreflang = [];
+  for (const m of html.matchAll(/<link[^>]+rel=["']alternate["'][^>]*>/gi)) {
+    const lang = m[0].match(/hreflang=["']([^"']*)["']/i)?.[1];
+    const href = m[0].match(/href=["']([^"']*)["']/i)?.[1];
+    if (lang && href) hreflang.push({ lang, href });
+  }
+  const ogTag = (p) => html.match(new RegExp(`<meta[^>]+property=["']og:${p}["'][^>]+content=["']([^"']*)["']`, "i"))?.[1]?.trim();
+  const og = ogTag("title") || ogTag("description") || ogTag("image") || ogTag("type") ? { title: ogTag("title"), description: ogTag("description"), image: ogTag("image"), type: ogTag("type") } : null;
+  const twTag = (p) => html.match(new RegExp(`<meta[^>]+name=["']twitter:${p}["'][^>]+content=["']([^"']*)["']`, "i"))?.[1]?.trim();
+  const twitter = twTag("card") || twTag("title") || twTag("description") ? { card: twTag("card"), title: twTag("title"), description: twTag("description") } : null;
+  const ampHref = html.match(/<link[^>]+rel=["']amphtml["'][^>]+href=["']([^"']*)["']/i)?.[1]?.trim() ?? null;
+  let imageCount = 0;
+  let imagesMissingAlt = 0;
+  const imageLinkSet = /* @__PURE__ */ new Set();
+  for (const m of html.matchAll(/<img\b([^>]*)>/gi)) {
+    imageCount++;
+    const alt = m[1].match(/\balt\s*=\s*["']([^"']*)["']/i)?.[1];
+    if (!alt || !alt.trim()) imagesMissingAlt++;
+    const src = m[1].match(/\bsrc\s*=\s*["']([^"']+)["']/i)?.[1];
+    if (src) {
+      try {
+        const abs = new URL(src, url);
+        if (abs.protocol === "http:" || abs.protocol === "https:") imageLinkSet.add(abs.href);
+      } catch {
+      }
+    }
+  }
+  if (og?.image) {
+    try {
+      imageLinkSet.add(new URL(og.image, url).href);
+    } catch {
+    }
+  }
+  const imageLinks = [...imageLinkSet].slice(0, 300);
+  const canonNorm = (u) => {
+    try {
+      const x = new URL(u, url);
+      return (x.origin + x.pathname).replace(/\/+$/, "");
+    } catch {
+      return u.replace(/\/+$/, "");
+    }
+  };
+  const canonicalSelf = canonicalUrl ? canonNorm(canonicalUrl) === canonNorm(url) : true;
+  let indexable = true;
+  let indexabilityReason = null;
+  if (status !== 200) {
+    indexable = false;
+    indexabilityReason = "non-200";
+  } else if (metaRobots?.includes("noindex")) {
+    indexable = false;
+    indexabilityReason = "noindex";
+  } else if (xRobotsTag?.toLowerCase().includes("noindex")) {
+    indexable = false;
+    indexabilityReason = "x-robots-noindex";
+  } else if (canonicalUrl && !canonicalSelf) {
+    indexable = false;
+    indexabilityReason = "canonicalised";
+  }
+  return {
+    url,
+    status,
+    via,
+    contentType,
+    responseTimeMs,
+    sizeBytes,
+    redirectUrl,
+    lastModified,
+    xRobotsTag,
+    title,
+    titleLength,
+    titlePixels,
+    metaDescription,
+    metaDescLength,
+    metaKeywords,
+    h1,
+    h1_2,
+    h2Count,
+    headings,
+    wordCount: wordCount2,
+    textRatio,
+    contentHash,
+    metaRobots,
+    relNext,
+    relPrev,
+    hreflang,
+    og,
+    twitter,
+    ampHref,
+    imageCount,
+    imagesMissingAlt,
+    imageLinks,
+    indexable,
+    indexabilityReason,
+    inSitemap: null,
+    schemaTypes,
+    canonicalUrl,
+    internalLinks,
+    externalLinks,
+    outlinks,
+    bodyMarkdown,
+    schema
+  };
+}
+async function extractPagesRotating(urls, opts) {
+  const fetched = await crawlWithRotation(urls, { apiKey: opts.kernelApiKey, concurrency: opts.concurrency, urlsPerBrowser: opts.urlsPerBrowser });
+  return fetched.map((r) => r.html ? parsePageData(r.url, r.html, r.status || 200, "browser", { headers: r.headers, responseTimeMs: r.responseTimeMs, redirectUrl: r.redirectUrl }) : emptyPageData(r.url, r.status || null, "browser"));
+}
+async function discoverUrlsWithSitemap(startUrl, maxPages, kernelApiKey) {
+  const sitemapUrls = await discoverSitemapUrls2(startUrl, maxPages);
+  if (sitemapUrls.length > 0) return { urls: sitemapUrls, sitemapUrls };
+  const spider = await spiderSite({ startUrl, maxUrls: maxPages, concurrency: 12, kernelApiKey });
+  return { urls: spider.urls.filter((u) => u.status === 200 || u.status === null).slice(0, maxPages).map((u) => u.url), sitemapUrls: spider.sitemapUrls };
+}
+async function fetchAndParse(url, kernelApiKey) {
+  const startedAt = Date.now();
+  try {
+    const res = await fetch(url, {
+      headers: { "User-Agent": UA2, "Accept": "text/html,application/xhtml+xml" },
+      signal: AbortSignal.timeout(1e4),
+      redirect: "manual"
+    });
+    const meta = {
+      headers: Object.fromEntries([...res.headers.entries()].map(([k, v]) => [k.toLowerCase(), v])),
+      responseTimeMs: Date.now() - startedAt,
+      redirectUrl: res.headers.get("location")
+    };
+    if (res.ok) {
+      const ct = res.headers.get("content-type") ?? "";
+      if (ct.includes("html")) {
+        const html = await res.text();
+        return parsePageData(url, html, res.status, "fetch", meta);
+      }
+      return { ...emptyPageData(url, res.status, "fetch"), contentType: ct || null, responseTimeMs: meta.responseTimeMs ?? null };
+    }
+    if ((res.status === 403 || res.status === 429) && kernelApiKey) {
+      const html = await fetchWithKernel(url);
+      return parsePageData(url, html, 200, "browser");
+    }
+    return { ...emptyPageData(url, res.status, "fetch"), responseTimeMs: meta.responseTimeMs ?? null, redirectUrl: meta.redirectUrl ?? null };
+  } catch {
+    if (kernelApiKey) {
+      try {
+        const html = await fetchWithKernel(url);
+        return parsePageData(url, html, 200, "browser");
+      } catch {
+      }
+    }
+    return emptyPageData(url, null, "fetch");
+  }
+}
+async function runWithConcurrency(items, concurrency, fn) {
+  let i = 0;
+  const workers = Array.from({ length: concurrency }, async () => {
+    while (i < items.length) {
+      const item = items[i++];
+      await fn(item);
+    }
+  });
+  await Promise.all(workers);
+}
+async function extractSite(opts) {
+  const startMs = Date.now();
+  const maxPages = Math.min(opts.maxPages ?? 100, 1e4);
+  const concurrency = Math.min(opts.concurrency ?? EXTRACT_CONCURRENCY, 10);
+  const rotating = Boolean(opts.rotateProxyEvery && opts.kernelApiKey);
+  const seeded = opts.seedUrls && opts.seedUrls.length > 0;
+  const pages = [];
+  let browserRetries = 0;
+  let truncated = false;
+  let spider;
+  let sitemapUrls = [];
+  if (rotating) {
+    const browserConcurrency = Math.max(1, Math.min(opts.parallelism ?? 1, 8));
+    const urlsPerBrowser = opts.rotateProxyEvery ?? 10;
+    const followLinks = !seeded;
+    const seen = /* @__PURE__ */ new Set();
+    const frontier = [];
+    const enqueue = (u) => {
+      const n = normalizeUrl(u, opts.startUrl);
+      if (n && !seen.has(n)) {
+        seen.add(n);
+        frontier.push(n);
+      }
+    };
+    if (seeded) {
+      for (const u of opts.seedUrls) enqueue(u);
+    } else {
+      enqueue(opts.startUrl);
+      sitemapUrls = await discoverSitemapUrls2(opts.startUrl, maxPages);
+      for (const u of sitemapUrls) enqueue(u);
+    }
+    const waveCap = Math.max(browserConcurrency * urlsPerBrowser, 30);
+    let done = 0;
+    while (frontier.length > 0 && pages.length < maxPages) {
+      const capacity = maxPages - pages.length;
+      const wave = frontier.splice(0, Math.min(capacity, waveCap));
+      const fetched = await crawlWithRotation(wave, {
+        apiKey: opts.kernelApiKey,
+        concurrency: browserConcurrency,
+        urlsPerBrowser,
+        onPage: (p) => {
+          done++;
+          opts.onProgress?.({ done, discovered: seen.size, frontier: frontier.length, lastOk: p.ok ? 1 : 0, lastFail: p.ok ? 0 : 1, phase: p.phase });
+        }
+      });
+      for (const r of fetched) {
+        const page = r.html ? parsePageData(r.url, r.html, r.status || 200, "browser", { headers: r.headers, responseTimeMs: r.responseTimeMs, redirectUrl: r.redirectUrl }) : emptyPageData(r.url, r.status || null, "browser");
+        pages.push(page);
+        browserRetries++;
+        if (followLinks && r.html) {
+          for (const l of page.outlinks) if (l.internal) enqueue(l.href);
+        }
+      }
+    }
+    truncated = frontier.length > 0;
+    spider = makeSeedSpider(opts.startUrl, [...seen]);
+  } else {
+    spider = seeded ? makeSeedSpider(opts.startUrl, opts.seedUrls) : await spiderSite({ startUrl: opts.startUrl, maxUrls: maxPages, concurrency: 12, kernelApiKey: opts.kernelApiKey });
+    sitemapUrls = spider.sitemapUrls;
+    const urlsToExtract = spider.urls.filter((u) => u.status === 200 || u.status === null).slice(0, maxPages).map((u) => u.url);
+    browserRetries = spider.browserRetries;
+    await runWithConcurrency(urlsToExtract, concurrency, async (url) => {
+      const data = await fetchAndParse(url, opts.kernelApiKey);
+      if (data.via === "browser") browserRetries++;
+      pages.push(data);
+    });
+    truncated = spider.truncated || pages.length >= maxPages;
+  }
+  const sitemapSet = sitemapUrls.length > 0 ? new Set(sitemapUrls.map((u) => normalizeUrl(u, opts.startUrl) ?? u)) : null;
+  for (const p of pages) p.inSitemap = sitemapSet ? sitemapSet.has(normalizeUrl(p.url, opts.startUrl) ?? p.url) : null;
+  let branding = null;
+  if (opts.formats?.includes("branding") && opts.kernelApiKey) {
+    branding = await extractBranding(opts.startUrl, opts.kernelApiKey).catch(() => null);
+  }
+  const seoAudit = opts.formats?.includes("issues") ? buildSeoAudit(pages, opts.startUrl) : null;
+  let imageAudit = null;
+  if (opts.formats?.includes("images")) {
+    imageAudit = await auditImages(pages, { concurrency: 12, timeoutMs: 12e3 });
+  }
+  return {
+    startUrl: opts.startUrl,
+    pages,
+    totalPages: pages.length,
+    spider,
+    durationMs: Date.now() - startMs,
+    truncated,
+    browserRetries,
+    branding,
+    seoAudit,
+    imageAudit
+  };
+}
+var import_turndown2, import_node_crypto3, PIXEL_WIDTHS, UA2, EXTRACT_CONCURRENCY, MAX_PAGE_MARKDOWN, turndown;
+var init_site_extractor = __esm({
+  "src/api/site-extractor.ts"() {
+    "use strict";
+    import_turndown2 = __toESM(require("turndown"), 1);
+    import_node_crypto3 = require("crypto");
+    init_site_mapper();
+    init_kernel_fetch();
+    init_rotating_proxy_crawl();
+    init_screenshot();
+    init_image_audit();
+    init_seo_audit();
+    PIXEL_WIDTHS = { i: 4, l: 4, j: 4, ".": 4, ",": 4, "'": 4, t: 6, f: 6, r: 6, " ": 4, m: 14, w: 13, W: 16, M: 16 };
+    UA2 = "Mozilla/5.0 (compatible; ThorbitBot/1.0; +https://thorbit.ai)";
+    EXTRACT_CONCURRENCY = 6;
+    MAX_PAGE_MARKDOWN = 4e4;
+    turndown = new import_turndown2.default({ headingStyle: "atx", bulletListMarker: "-", codeBlockStyle: "fenced" });
+  }
+});
+
+// src/inngest/client.ts
+var import_inngest, inngest;
+var init_client = __esm({
+  "src/inngest/client.ts"() {
+    "use strict";
+    import_inngest = require("inngest");
+    inngest = new import_inngest.Inngest({ id: "mcp-scraper" });
+  }
+});
+
+// src/lib/llm-parse-with-retry.ts
+function extractJson(raw) {
+  let text = raw.replace(/<think>[\s\S]*?<\/think>/gi, "").trim();
+  text = text.replace(/^```(?:json)?\s*/i, "").replace(/\s*```\s*$/, "").trim();
+  if (text.startsWith("{") || text.startsWith("[")) return text;
+  const objStart = text.indexOf("{");
+  const arrStart = text.indexOf("[");
+  let start = -1;
+  let open, close;
+  if (objStart === -1 && arrStart === -1) return text;
+  if (objStart === -1) {
+    start = arrStart;
+    open = "[";
+    close = "]";
+  } else if (arrStart === -1) {
+    start = objStart;
+    open = "{";
+    close = "}";
+  } else if (arrStart < objStart) {
+    start = arrStart;
+    open = "[";
+    close = "]";
+  } else {
+    start = objStart;
+    open = "{";
+    close = "}";
+  }
+  let depth = 0;
+  let inStr = false;
+  let escape = false;
+  for (let i = start; i < text.length; i++) {
+    const ch = text[i];
+    if (escape) {
+      escape = false;
+      continue;
+    }
+    if (ch === "\\" && inStr) {
+      escape = true;
+      continue;
+    }
+    if (ch === '"') {
+      inStr = !inStr;
+      continue;
+    }
+    if (inStr) continue;
+    if (ch === open) depth++;
+    else if (ch === close) {
+      if (--depth === 0) return text.slice(start, i + 1);
+    }
+  }
+  return text.slice(start);
+}
+async function llmParseWithRetry(schema, promptFn, llm) {
+  let lastError;
+  for (let attempt = 0; attempt < 3; attempt++) {
+    const prompt = attempt === 0 ? promptFn(attempt) : promptFn(attempt, lastError?.message);
+    let raw;
+    try {
+      raw = await llm.completeJson(prompt);
+    } catch (e) {
+      lastError = e instanceof Error ? e : new Error(String(e));
+      continue;
+    }
+    const result = schema.safeParse(raw);
+    if (result.success) return result.data;
+    lastError = result.error;
+  }
+  const detail = lastError?.message?.slice(0, 200) ?? "unknown error";
+  throw new Error(`LLM failed to produce valid output after 3 attempts: ${detail}`);
+}
+var import_sdk4, DeepInfraLlmClient, OpenRouterLlmClient, FallbackLlmClient;
+var init_llm_parse_with_retry = __esm({
+  "src/lib/llm-parse-with-retry.ts"() {
+    "use strict";
+    import_sdk4 = __toESM(require("@anthropic-ai/sdk"), 1);
+    init_cost_telemetry();
+    DeepInfraLlmClient = class {
+      apiKey;
+      model;
+      totalCostUsd = 0;
+      constructor(apiKey, model = "Qwen/Qwen3.6-35B-A3B") {
+        this.apiKey = apiKey;
+        this.model = model;
+      }
+      async complete(prompt) {
+        const resp = await fetch("https://api.deepinfra.com/v1/openai/chat/completions", {
+          method: "POST",
+          headers: { "Content-Type": "application/json", "Authorization": `Bearer ${this.apiKey}` },
+          body: JSON.stringify({ model: this.model, messages: [{ role: "user", content: prompt }], max_tokens: 8192 })
+        });
+        if (!resp.ok) throw new Error(`DeepInfra error ${resp.status}: ${await resp.text()}`);
+        const data = await resp.json();
+        void recordVendorUsage({ vendor: "deepinfra_qwen", model: this.model, units: data.usage?.completion_tokens ?? 0, unitType: "tokens" });
+        return data.choices[0].message.content;
+      }
+      async completeJson(prompt) {
+        const raw = await this.complete(prompt);
+        try {
+          return JSON.parse(extractJson(raw));
+        } catch {
+          const coerced = await this.complete(
+            `Convert the following text to a valid JSON object. Output ONLY the JSON, no markdown, no explanation:
+
+${raw}`
+          );
+          return JSON.parse(extractJson(coerced));
+        }
+      }
+    };
+    OpenRouterLlmClient = class {
+      apiKey;
+      model;
+      totalCostUsd = 0;
+      constructor(apiKey, model = "qwen/qwen3-235b-a22b") {
+        this.apiKey = apiKey;
+        this.model = model;
+      }
+      async complete(prompt) {
+        const resp = await fetch("https://openrouter.ai/api/v1/chat/completions", {
+          method: "POST",
+          headers: { "Content-Type": "application/json", "Authorization": `Bearer ${this.apiKey}` },
+          body: JSON.stringify({ model: this.model, messages: [{ role: "user", content: prompt }], max_tokens: 8192 })
+        });
+        if (!resp.ok) throw new Error(`OpenRouter error ${resp.status}: ${await resp.text()}`);
+        const data = await resp.json();
+        const cost = data.usage?.cost ?? 0;
+        this.totalCostUsd += cost;
+        void recordVendorUsage({ vendor: "openrouter", model: this.model, units: cost, unitType: "usd" });
+        return data.choices[0].message.content;
+      }
+      async completeJson(prompt) {
+        const raw = await this.complete(prompt);
+        try {
+          return JSON.parse(extractJson(raw));
+        } catch {
+          const coerced = await this.complete(
+            `Convert the following text to a valid JSON object. Output ONLY the JSON, no markdown, no explanation:
+
+${raw}`
+          );
+          return JSON.parse(extractJson(coerced));
+        }
+      }
+    };
+    FallbackLlmClient = class {
+      constructor(primary, secondary) {
+        this.primary = primary;
+        this.secondary = secondary;
+      }
+      primary;
+      secondary;
+      lastServedBy = null;
+      lastPrimaryError = null;
+      get totalCostUsd() {
+        return this.primary.totalCostUsd + this.secondary.totalCostUsd;
+      }
+      async complete(prompt) {
+        try {
+          const result = await this.primary.complete(prompt);
+          this.lastServedBy = "primary";
+          this.lastPrimaryError = null;
+          return result;
+        } catch (err) {
+          this.lastPrimaryError = String(err);
+          console.warn("[FallbackLlmClient] primary failed, using secondary:", err);
+          this.lastServedBy = "secondary";
+          return this.secondary.complete(prompt);
+        }
+      }
+      async completeJson(prompt) {
+        try {
+          const result = await this.primary.completeJson(prompt);
+          this.lastServedBy = "primary";
+          this.lastPrimaryError = null;
+          return result;
+        } catch (err) {
+          this.lastPrimaryError = String(err);
+          console.warn("[FallbackLlmClient] primary failed, using secondary:", err);
+          this.lastServedBy = "secondary";
+          return this.secondary.completeJson(prompt);
+        }
+      }
+    };
+  }
+});
+
+// src/lib/http-client.ts
+var FetchHttpClient;
+var init_http_client = __esm({
+  "src/lib/http-client.ts"() {
+    "use strict";
+    FetchHttpClient = class {
+      async get(url, headers) {
+        const response = await fetch(url, { headers });
+        const body = await response.text();
+        return { status: response.status, body };
+      }
+      async post(url, body, headers) {
+        const response = await fetch(url, {
+          method: "POST",
+          headers: { "Content-Type": "application/json", ...headers },
+          body: JSON.stringify(body)
+        });
+        const responseBody = await response.text();
+        return { status: response.status, body: responseBody };
+      }
+    };
+  }
+});
+
+// src/lib/site-architecture-auditor/rules.ts
+function applyGuardrailRules(raw, context) {
+  const passed = [];
+  const flagged = [];
+  for (const item of raw) {
+    const flags = [];
+    const rec = item;
+    const anchor = typeof rec["anchor_text"] === "string" ? rec["anchor_text"].trim().toLowerCase() : null;
+    if (anchor !== null) {
+      const isVague = VAGUE_ANCHOR_BLOCKLIST.some(
+        (blocked) => blocked.toLowerCase() === anchor
+      );
+      if (isVague) flags.push("vague_anchor");
+    }
+    const outlinks = typeof rec["outlink_count"] === "number" ? rec["outlink_count"] : null;
+    if (outlinks !== null && outlinks >= EXCESSIVE_OUTLINKS_THRESHOLD) {
+      flags.push("excessive_outlinks");
+    }
+    if (context !== null && context !== void 0 && typeof context["resolveFlags"] === "function") {
+      const resolveFlags = context["resolveFlags"];
+      const contextualCandidates = [
+        "seo_only_link",
+        "exact_match_repetition",
+        "orphaned_support_page",
+        "crawl_depth_regression",
+        "low_content_source",
+        "circular_authority_trap"
+      ];
+      const contextFlags = resolveFlags(item, contextualCandidates);
+      flags.push(...contextFlags);
+    }
+    if (flags.length === 0) {
+      passed.push(item);
+    } else {
+      flagged.push({ item, flags });
+    }
+  }
+  return { passed, flagged };
+}
+var import_zod2, PageTypeSchema, ClassificationConfidenceSchema, ARCHITECTURE_SCORE_WEIGHTS, ILES_WEIGHTS, PriorityTierSchema, EXCESSIVE_OUTLINKS_THRESHOLD, VAGUE_ANCHOR_BLOCKLIST, GuardrailFlagNameSchema, RequiredReportSectionSchema, LocationArchitecturePatternSchema;
+var init_rules = __esm({
+  "src/lib/site-architecture-auditor/rules.ts"() {
+    "use strict";
+    import_zod2 = require("zod");
+    PageTypeSchema = import_zod2.z.enum([
+      "homepage",
+      "core_service",
+      "sub_service",
+      "location",
+      "service_location",
+      "industry",
+      "product",
+      "pricing",
+      "faq",
+      "blog_post",
+      "guide_resource",
+      "case_study",
+      "comparison",
+      "competitor",
+      "about",
+      "contact",
+      "team",
+      "careers",
+      "privacy_policy",
+      "terms",
+      "html_sitemap",
+      "login_portal",
       "support_help_center"
     ]);
     ClassificationConfidenceSchema = import_zod2.z.enum([
@@ -9602,883 +10079,449 @@ var init_site_audit_service = __esm({
                 topic_cluster: null,
                 location_signals: c.locationSignals
               });
-            }
-            if (unclassifiedUrls.length > 0) {
-              const unclassifiedBatch = batch.filter((u) => unclassifiedUrls.includes(u.url));
-              const llmResults = await llmParseWithRetry(
-                import_zod10.z.array(PageTypeClassificationRowSchema),
-                (attempt, error) => {
-                  void attempt;
-                  void error;
-                  return buildMetricsClassifyPrompt({
-                    urls: unclassifiedBatch,
-                    siteDomain: clientDomain
-                  });
-                },
-                this.deps.llm
-              );
-              allClassifications.push(...llmResults);
-            }
-          }
-        });
-        void sessionPath;
-        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase3-classify", {
-          classifications: allClassifications,
-          totalClassified: allClassifications.length
-        });
-      }
-      async runComparePhase(jobId, payload) {
-        void payload;
-        const job = await this.deps.repo.getSiteAuditJob(jobId);
-        if (!job) throw new Error(`[site-audit-service] runComparePhase: job ${jobId} not found`);
-        const request = JSON.parse(job.request);
-        const clientDomain = request.clientDomain;
-        const result = await this.runLlmBilled(
-          job.user_id,
-          `${jobId}:phase4-compare`,
-          () => llmParseWithRetry(
-            CompareRecommendOutputSchema,
-            (attempt, error) => {
-              void attempt;
-              void error;
-              return buildCompareRecommendPrompt({
-                clientDomain,
-                clientMoneyPages: [],
-                competitorSites: [],
-                clientUrlMetrics: [],
-                clientPageTypes: []
-              });
-            },
-            this.deps.llm
-          )
-        );
-        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase4-compare", result);
-        return result;
-      }
-      async runSynthesizePhase(jobId, payload) {
-        void payload;
-        const job = await this.deps.repo.getSiteAuditJob(jobId);
-        if (!job) throw new Error(`[site-audit-service] runSynthesizePhase: job ${jobId} not found`);
-        const request = JSON.parse(job.request);
-        const clientDomain = request.clientDomain;
-        const sessionPath = request.sessionPath;
-        const architectureScoreComponents = {};
-        for (const [dim, weight] of Object.entries(ARCHITECTURE_SCORE_WEIGHTS)) {
-          architectureScoreComponents[dim] = weight * 50;
-        }
-        const architectureTotal = Object.values(architectureScoreComponents).reduce((s, v) => s + v, 0);
-        const ilesComponents = {};
-        for (const [dim, weight] of Object.entries(ILES_WEIGHTS)) {
-          ilesComponents[dim] = weight * 50;
-        }
-        const ilesTotal = Object.values(ilesComponents).reduce((s, v) => s + v, 0);
-        const preComputedArchitectureScore = {
-          total: architectureTotal,
-          components: architectureScoreComponents
-        };
-        const preComputedIlesScore = {
-          total: ilesTotal,
-          components: ilesComponents
-        };
-        const reportMarkdown = await this.runLlmBilled(
-          job.user_id,
-          `${jobId}:phase5-synthesize`,
-          () => llmParseWithRetry(
-            import_zod10.z.string(),
-            (attempt, error) => {
-              void attempt;
-              void error;
-              return buildScoreSynthesizePrompt({
-                preComputedArchitectureScore,
-                preComputedIlesScore,
-                rankedBacklog: [],
-                clientDomain,
-                sessionPath
-              });
-            },
-            this.deps.llm
-          )
-        );
-        const auditResult = SiteAuditResultSchema.parse({
-          jobId,
-          clientDomain,
-          architectureScore: preComputedArchitectureScore,
-          ilesScore: preComputedIlesScore,
-          rankedBacklog: [],
-          linkRecommendations: [],
-          newPagesRecommended: [],
-          guardrailFlags: [],
-          reportMarkdown
-        });
-        await this.deps.repo.completeSiteAuditJob(jobId, auditResult);
-        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase5-synthesize", {
-          architectureScore: auditResult.architectureScore,
-          ilesScore: auditResult.ilesScore
-        });
-      }
-      dispatchPhase(_phase, _jobId, _payload) {
-        throw new Error("dispatchPhase delegates to phases.ts \u2014 see saa-011");
-      }
-      async _readCompetitorDomains(request) {
-        try {
-          const response = await this.deps.http.get(`file://${request.competitorDomainsCsvPath}`);
-          const lines = response.body.split("\n").map((l) => l.trim()).filter((l) => l.length > 0 && !l.startsWith("#"));
-          const domains = [];
-          for (const line of lines) {
-            if (line.includes(",")) {
-              domains.push(...line.split(",").map((d) => d.trim()).filter((d) => d.length > 0));
-            } else {
-              domains.push(line);
-            }
-          }
-          return domains.filter((d) => d.includes("."));
-        } catch {
-          return [];
-        }
-      }
-    };
-  }
-});
-
-// src/services/site-architecture-auditor/site-audit-repository.ts
-function rowToSiteAuditJobRow(row) {
-  return SiteAuditJobRowSchema.parse({
-    id: String(row.id),
-    user_id: Number(row.user_id),
-    status: row.status != null ? String(row.status) : "pending",
-    client_domain: String(row.client_domain),
-    session_path: String(row.session_path),
-    request: String(row.request),
-    result: row.result != null ? String(row.result) : null,
-    error: row.error != null ? String(row.error) : null,
-    created_at: String(row.created_at),
-    updated_at: String(row.updated_at)
-  });
-}
-var SiteAuditRepository;
-var init_site_audit_repository = __esm({
-  "src/services/site-architecture-auditor/site-audit-repository.ts"() {
-    "use strict";
-    init_db();
-    init_db();
-    SiteAuditRepository = class {
-      db;
-      constructor() {
-        this.db = getDb();
-      }
-      async createSiteAuditJob(jobId, userId, request) {
-        await this.db.execute({
-          sql: `INSERT INTO site_audit_jobs
-              (id, user_id, status, client_domain, session_path, request, created_at, updated_at)
-            VALUES (?, ?, 'pending', ?, ?, ?, datetime('now'), datetime('now'))`,
-          args: [
-            jobId,
-            userId,
-            request.clientDomain,
-            request.sessionPath,
-            JSON.stringify(request)
-          ]
-        });
-      }
-      async claimPendingSiteAuditJob() {
-        const upd = await this.db.execute(
-          `UPDATE site_audit_jobs
-         SET status = 'running', updated_at = datetime('now')
-       WHERE id = (
-         SELECT id FROM site_audit_jobs
-         WHERE status = 'pending'
-         ORDER BY created_at ASC
-         LIMIT 1
-       )
-       RETURNING *`
-        );
-        if (upd.rowsAffected === 0) return null;
-        const raw = upd.rows[0];
-        if (!raw) return null;
-        return rowToSiteAuditJobRow(raw);
-      }
-      async getSiteAuditJob(jobId) {
-        const res = await this.db.execute({
-          sql: `SELECT * FROM site_audit_jobs WHERE id = ?`,
-          args: [jobId]
-        });
-        const raw = res.rows[0];
-        if (!raw) return null;
-        return rowToSiteAuditJobRow(raw);
-      }
-      async listSiteAuditJobs(userId) {
-        const res = userId != null ? await this.db.execute({
-          sql: `SELECT * FROM site_audit_jobs WHERE user_id = ? ORDER BY created_at DESC LIMIT 50`,
-          args: [userId]
-        }) : await this.db.execute(
-          `SELECT * FROM site_audit_jobs ORDER BY created_at DESC LIMIT 50`
-        );
-        return res.rows.map((r) => rowToSiteAuditJobRow(r));
-      }
-      async updateSiteAuditJobState(jobId, status) {
-        await this.db.execute({
-          sql: `UPDATE site_audit_jobs SET status = ?, updated_at = datetime('now') WHERE id = ?`,
-          args: [status, jobId]
-        });
-      }
-      async completeSiteAuditJob(jobId, result) {
-        await this.db.execute({
-          sql: `UPDATE site_audit_jobs
-              SET status = 'complete', result = ?, updated_at = datetime('now')
-            WHERE id = ?`,
-          args: [JSON.stringify(result), jobId]
-        });
-      }
-      async failSiteAuditJob(jobId, error) {
-        await this.db.execute({
-          sql: `UPDATE site_audit_jobs
-              SET status = 'failed', error = ?, updated_at = datetime('now')
-            WHERE id = ?`,
-          args: [error, jobId]
+            }
+            if (unclassifiedUrls.length > 0) {
+              const unclassifiedBatch = batch.filter((u) => unclassifiedUrls.includes(u.url));
+              const llmResults = await llmParseWithRetry(
+                import_zod10.z.array(PageTypeClassificationRowSchema),
+                (attempt, error) => {
+                  void attempt;
+                  void error;
+                  return buildMetricsClassifyPrompt({
+                    urls: unclassifiedBatch,
+                    siteDomain: clientDomain
+                  });
+                },
+                this.deps.llm
+              );
+              allClassifications.push(...llmResults);
+            }
+          }
         });
-      }
-      async logSiteAuditPhaseComplete(jobId, phase, output) {
-        await this.db.execute({
-          sql: `INSERT OR IGNORE INTO site_audit_phase_log (job_id, phase, output_summary, completed_at)
-            VALUES (?, ?, ?, datetime('now'))`,
-          args: [jobId, phase, JSON.stringify(output)]
-        }).catch((err) => {
-          console.warn(
-            "[site-audit-repository] logSiteAuditPhaseComplete failed:",
-            err instanceof Error ? err.message : String(err)
-          );
+        void sessionPath;
+        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase3-classify", {
+          classifications: allClassifications,
+          totalClassified: allClassifications.length
         });
       }
-    };
-  }
-});
-
-// src/services/site-architecture-auditor/python/graph-runner.ts
-var childProcess, path, util, import_node_url, import_meta, execFile2, __dirname, SiteAuditPythonRunner;
-var init_graph_runner = __esm({
-  "src/services/site-architecture-auditor/python/graph-runner.ts"() {
-    "use strict";
-    childProcess = __toESM(require("child_process"), 1);
-    path = __toESM(require("path"), 1);
-    util = __toESM(require("util"), 1);
-    import_node_url = require("url");
-    init_schemas();
-    import_meta = {};
-    execFile2 = util.promisify(childProcess.execFile);
-    __dirname = path.dirname((0, import_node_url.fileURLToPath)(import_meta.url));
-    SiteAuditPythonRunner = class {
-      graphMetricsScript;
-      locationClassifierScript;
-      constructor() {
-        try {
-          childProcess.execFileSync("python3", ["-c", "import networkx"]);
-        } catch {
-          throw new Error("SiteAuditPythonRunner: networkx not available");
-        }
-        this.graphMetricsScript = path.join(__dirname, "scripts", "compute_graph_metrics.py");
-        this.locationClassifierScript = path.join(__dirname, "scripts", "location_classifier.py");
-      }
-      async runGraphMetrics(input) {
-        const validatedInput = PythonGraphInputSchema.parse(input);
-        let stdout;
-        let stderr;
-        try {
-          ({ stdout, stderr } = await execFile2(
-            "python3",
-            [this.graphMetricsScript, "--input", JSON.stringify(validatedInput)],
-            { timeout: 12e4 }
-          ));
-        } catch (err) {
-          const execError = err;
-          throw new Error(`python-subprocess-001: ${execError.stderr ?? String(err)}`);
-        }
-        void stderr;
-        return PythonGraphOutputSchema.parse(JSON.parse(stdout));
-      }
-      async runLocationClassifier(input) {
-        const validatedInput = LocationClassifierInputSchema.parse(input);
-        let stdout;
-        let stderr;
-        try {
-          ({ stdout, stderr } = await execFile2(
-            "python3",
-            [this.locationClassifierScript, "--input", JSON.stringify(validatedInput)],
-            { timeout: 12e4 }
-          ));
-        } catch (err) {
-          const execError = err;
-          throw new Error(`python-subprocess-001: ${execError.stderr ?? String(err)}`);
-        }
-        void stderr;
-        return LocationClassifierOutputSchema.parse(JSON.parse(stdout));
+      async runComparePhase(jobId, payload) {
+        void payload;
+        const job = await this.deps.repo.getSiteAuditJob(jobId);
+        if (!job) throw new Error(`[site-audit-service] runComparePhase: job ${jobId} not found`);
+        const request = JSON.parse(job.request);
+        const clientDomain = request.clientDomain;
+        const result = await this.runLlmBilled(
+          job.user_id,
+          `${jobId}:phase4-compare`,
+          () => llmParseWithRetry(
+            CompareRecommendOutputSchema,
+            (attempt, error) => {
+              void attempt;
+              void error;
+              return buildCompareRecommendPrompt({
+                clientDomain,
+                clientMoneyPages: [],
+                competitorSites: [],
+                clientUrlMetrics: [],
+                clientPageTypes: []
+              });
+            },
+            this.deps.llm
+          )
+        );
+        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase4-compare", result);
+        return result;
       }
-    };
-  }
-});
-
-// src/services/site-architecture-auditor/factory.ts
-function makeSiteAuditService() {
-  const deepInfraKey = process.env["DEEPINFRA_API_KEY"];
-  if (!deepInfraKey) throw new Error("DEEPINFRA_API_KEY is required");
-  const primary = new DeepInfraLlmClient(deepInfraKey);
-  const llm = process.env["OPENROUTER_API_KEY"] ? new FallbackLlmClient(primary, new OpenRouterLlmClient(process.env["OPENROUTER_API_KEY"])) : primary;
-  const repo = new SiteAuditRepository();
-  const http = new FetchHttpClient();
-  const python = new SiteAuditPythonRunner();
-  return new SiteAuditService({ repo, llm, http, python });
-}
-var init_factory = __esm({
-  "src/services/site-architecture-auditor/factory.ts"() {
-    "use strict";
-    init_llm_parse_with_retry();
-    init_http_client();
-    init_site_audit_service();
-    init_site_audit_repository();
-    init_graph_runner();
-  }
-});
-
-// src/services/site-architecture-auditor/phases.ts
-async function dispatchSiteAuditPhase(service, phase, raw) {
-  switch (phase) {
-    case "phase1-ingest":
-      return handleIngestPhase(service, raw);
-    case "phase2-build-graph":
-      return handleBuildGraphPhase(service, raw);
-    case "phase3-classify":
-      return handleClassifyPhase(service, raw);
-    case "phase4-compare":
-      return handleComparePhase(service, raw);
-    case "phase5-synthesize":
-      return handleSynthesizePhase(service, raw);
-    default: {
-      const _exhaustive = phase;
-      throw new Error("Unknown phase: " + String(_exhaustive));
-    }
-  }
-}
-async function handleIngestPhase(service, raw) {
-  const payload = SiteAuditIngestPayloadSchema.parse(raw);
-  await service.runIngestPhase(payload.jobId, payload);
-  await service.runEnrichPhase(payload.jobId, payload);
-}
-async function handleBuildGraphPhase(service, raw) {
-  const payload = SiteAuditBuildGraphPayloadSchema.parse(raw);
-  const limit = (0, import_p_limit2.default)(3);
-  const tasks = [
-    limit(() => service.runBuildGraphPhase(payload.jobId, payload))
-  ];
-  const results = await Promise.allSettled(tasks);
-  for (const result of results) {
-    if (result.status === "rejected") {
-      console.warn(
-        "[phases] handleBuildGraphPhase: phase failed for job",
-        payload.jobId,
-        result.reason instanceof Error ? result.reason.message : String(result.reason)
-      );
-    }
-  }
-}
-async function handleClassifyPhase(service, raw) {
-  const payload = SiteAuditClassifyPayloadSchema.parse(raw);
-  const limit = (0, import_p_limit2.default)(3);
-  const tasks = [
-    limit(() => service.runClassifyPhase(payload.jobId, payload))
-  ];
-  const results = await Promise.allSettled(tasks);
-  for (const result of results) {
-    if (result.status === "rejected") {
-      console.warn(
-        "[phases] handleClassifyPhase: phase failed for job",
-        payload.jobId,
-        result.reason instanceof Error ? result.reason.message : String(result.reason)
-      );
-    }
-  }
-}
-async function handleComparePhase(service, raw) {
-  const payload = SiteAuditComparePayloadSchema.parse(raw);
-  const compareResult = await service.runComparePhase(payload.jobId, payload);
-  const rawRecommendations = compareResult.linkRecommendations ?? [];
-  const resolveFlagsStub = (_item, _candidates) => {
-    return [];
-  };
-  const guardrailResult = applyGuardrailRules(rawRecommendations, { resolveFlags: resolveFlagsStub });
-  console.warn(
-    "[phases] handleComparePhase: guardrail check complete \u2014 passed:",
-    guardrailResult.passed.length,
-    "flagged:",
-    guardrailResult.flagged.length,
-    "(context-dependent flags require site-graph state not yet exposed by runComparePhase)"
-  );
-}
-async function handleSynthesizePhase(service, raw) {
-  const payload = SiteAuditSynthesizePayloadSchema.parse(raw);
-  await service.runSynthesizePhase(payload.jobId, payload);
-}
-var import_p_limit2;
-var init_phases = __esm({
-  "src/services/site-architecture-auditor/phases.ts"() {
-    "use strict";
-    import_p_limit2 = __toESM(require("p-limit"), 1);
-    init_schemas();
-    init_rules();
-  }
-});
-
-// src/inngest/functions/site-audit.ts
-var siteAuditFn;
-var init_site_audit = __esm({
-  "src/inngest/functions/site-audit.ts"() {
-    "use strict";
-    init_client();
-    init_factory();
-    init_phases();
-    siteAuditFn = inngest.createFunction(
-      { id: "site-audit", triggers: [{ event: "site-architecture-auditor/audit.requested" }] },
-      async ({ event, step }) => {
-        const { phase, payload } = event.data;
-        await step.run("dispatch-phase", async () => {
-          const service = makeSiteAuditService();
-          await dispatchSiteAuditPhase(service, phase, payload);
+      async runSynthesizePhase(jobId, payload) {
+        void payload;
+        const job = await this.deps.repo.getSiteAuditJob(jobId);
+        if (!job) throw new Error(`[site-audit-service] runSynthesizePhase: job ${jobId} not found`);
+        const request = JSON.parse(job.request);
+        const clientDomain = request.clientDomain;
+        const sessionPath = request.sessionPath;
+        const architectureScoreComponents = {};
+        for (const [dim, weight] of Object.entries(ARCHITECTURE_SCORE_WEIGHTS)) {
+          architectureScoreComponents[dim] = weight * 50;
+        }
+        const architectureTotal = Object.values(architectureScoreComponents).reduce((s, v) => s + v, 0);
+        const ilesComponents = {};
+        for (const [dim, weight] of Object.entries(ILES_WEIGHTS)) {
+          ilesComponents[dim] = weight * 50;
+        }
+        const ilesTotal = Object.values(ilesComponents).reduce((s, v) => s + v, 0);
+        const preComputedArchitectureScore = {
+          total: architectureTotal,
+          components: architectureScoreComponents
+        };
+        const preComputedIlesScore = {
+          total: ilesTotal,
+          components: ilesComponents
+        };
+        const reportMarkdown = await this.runLlmBilled(
+          job.user_id,
+          `${jobId}:phase5-synthesize`,
+          () => llmParseWithRetry(
+            import_zod10.z.string(),
+            (attempt, error) => {
+              void attempt;
+              void error;
+              return buildScoreSynthesizePrompt({
+                preComputedArchitectureScore,
+                preComputedIlesScore,
+                rankedBacklog: [],
+                clientDomain,
+                sessionPath
+              });
+            },
+            this.deps.llm
+          )
+        );
+        const auditResult = SiteAuditResultSchema.parse({
+          jobId,
+          clientDomain,
+          architectureScore: preComputedArchitectureScore,
+          ilesScore: preComputedIlesScore,
+          rankedBacklog: [],
+          linkRecommendations: [],
+          newPagesRecommended: [],
+          guardrailFlags: [],
+          reportMarkdown
+        });
+        await this.deps.repo.completeSiteAuditJob(jobId, auditResult);
+        await this.deps.repo.logSiteAuditPhaseComplete(jobId, "phase5-synthesize", {
+          architectureScore: auditResult.architectureScore,
+          ilesScore: auditResult.ilesScore
         });
-        return { phase, status: "done" };
       }
-    );
-  }
-});
-
-// src/api/seo-link-graph.ts
-function normalize(u) {
-  return u.split("#")[0].replace(/\/$/, "");
-}
-function buildLinkGraph(pages, startUrl) {
-  const statusByUrl = /* @__PURE__ */ new Map();
-  for (const p of pages) statusByUrl.set(normalize(p.url), p.status);
-  const edges = [];
-  for (const p of pages) {
-    for (const l of p.outlinks ?? []) {
-      const nofollow = (l.rel ?? "").split(/\s+/).includes("nofollow");
-      edges.push({
-        from: p.url,
-        to: l.href,
-        anchor: l.anchor,
-        rel: l.rel,
-        internal: l.internal,
-        nofollow,
-        targetStatus: statusByUrl.get(normalize(l.href)) ?? null
-      });
-    }
-  }
-  const inboundFrom = /* @__PURE__ */ new Map();
-  const inboundAnchors = /* @__PURE__ */ new Map();
-  const adjacency = /* @__PURE__ */ new Map();
-  for (const e of edges) {
-    if (!e.internal) continue;
-    const to = normalize(e.to);
-    const from = normalize(e.from);
-    if (!inboundFrom.has(to)) inboundFrom.set(to, /* @__PURE__ */ new Set());
-    inboundFrom.get(to).add(from);
-    if (e.anchor) {
-      if (!inboundAnchors.has(to)) inboundAnchors.set(to, []);
-      inboundAnchors.get(to).push(e.anchor);
-    }
-    if (!e.nofollow && (statusByUrl.get(to) === 200 || !statusByUrl.has(to))) {
-      if (!adjacency.has(from)) adjacency.set(from, /* @__PURE__ */ new Set());
-      adjacency.get(from).add(to);
-    }
-  }
-  const depth = /* @__PURE__ */ new Map();
-  const start = normalize(startUrl);
-  const queue = [start];
-  depth.set(start, 0);
-  while (queue.length > 0) {
-    const cur = queue.shift();
-    const d = depth.get(cur);
-    for (const next of adjacency.get(cur) ?? []) {
-      if (!depth.has(next)) {
-        depth.set(next, d + 1);
-        queue.push(next);
+      dispatchPhase(_phase, _jobId, _payload) {
+        throw new Error("dispatchPhase delegates to phases.ts \u2014 see saa-011");
       }
-    }
-  }
-  const topAnchorsFor = (url) => {
-    const counts = /* @__PURE__ */ new Map();
-    for (const a of inboundAnchors.get(url) ?? []) counts.set(a, (counts.get(a) ?? 0) + 1);
-    return [...counts.entries()].sort((a, b) => b[1] - a[1]).slice(0, 3).map((e) => e[0]);
-  };
-  const metrics = /* @__PURE__ */ new Map();
-  for (const p of pages) {
-    const key = normalize(p.url);
-    const inbound = inboundFrom.get(key) ?? /* @__PURE__ */ new Set();
-    metrics.set(p.url, {
-      url: p.url,
-      inlinks: inbound.size,
-      uniqueInlinks: inbound.size,
-      outlinksInternal: (p.outlinks ?? []).filter((l) => l.internal).length,
-      outlinksExternal: (p.outlinks ?? []).filter((l) => !l.internal).length,
-      crawlDepth: depth.has(key) ? depth.get(key) : null,
-      orphan: inbound.size === 0 && key !== start,
-      topAnchors: topAnchorsFor(key)
-    });
-  }
-  return { edges, metrics };
-}
-var init_seo_link_graph = __esm({
-  "src/api/seo-link-graph.ts"() {
-    "use strict";
+      async _readCompetitorDomains(request) {
+        try {
+          const response = await this.deps.http.get(`file://${request.competitorDomainsCsvPath}`);
+          const lines = response.body.split("\n").map((l) => l.trim()).filter((l) => l.length > 0 && !l.startsWith("#"));
+          const domains = [];
+          for (const line of lines) {
+            if (line.includes(",")) {
+              domains.push(...line.split(",").map((d) => d.trim()).filter((d) => d.length > 0));
+            } else {
+              domains.push(line);
+            }
+          }
+          return domains.filter((d) => d.includes("."));
+        } catch {
+          return [];
+        }
+      }
+    };
   }
 });
 
-// src/api/seo-link-report.ts
-function registrableDomain(host) {
-  const h = host.replace(/^www\./, "").toLowerCase();
-  const parts = h.split(".");
-  return parts.length <= 2 ? h : parts.slice(-2).join(".");
-}
-function buildLinkReport(edges, metrics, siteUrl) {
-  let siteReg = "";
-  try {
-    siteReg = registrableDomain(new URL(siteUrl).hostname);
-  } catch {
-    siteReg = "";
-  }
-  const internalEdges = edges.filter((e) => e.internal);
-  const pages = metrics.length;
-  const orphans = metrics.filter((m) => m.orphan).length;
-  const brokenInternal = internalEdges.filter((e) => e.targetStatus != null && e.targetStatus >= 400).length;
-  const sumInlinks = metrics.reduce((a, m) => a + m.inlinks, 0);
-  const sumOutlinks = metrics.reduce((a, m) => a + m.outlinksInternal + m.outlinksExternal, 0);
-  const distribution = { zero: 0, oneToTwo: 0, threeToTen: 0, elevenPlus: 0 };
-  for (const m of metrics) {
-    if (m.inlinks === 0) distribution.zero++;
-    else if (m.inlinks <= 2) distribution.oneToTwo++;
-    else if (m.inlinks <= 10) distribution.threeToTen++;
-    else distribution.elevenPlus++;
-  }
-  const topByInlinks = [...metrics].sort((a, b) => b.inlinks - a.inlinks).slice(0, 20).map((m) => ({ url: m.url, inlinks: m.inlinks, outlinksInternal: m.outlinksInternal, outlinksExternal: m.outlinksExternal }));
-  const domMap = /* @__PURE__ */ new Map();
-  let externalTotal = 0;
-  for (const e of edges) {
-    let domain;
-    try {
-      domain = registrableDomain(new URL(e.to).hostname);
-    } catch {
-      continue;
-    }
-    if (!domain || domain === siteReg) continue;
-    externalTotal++;
-    const d = domMap.get(domain) ?? { links: 0, nofollow: 0, pages: /* @__PURE__ */ new Set() };
-    d.links++;
-    if (e.nofollow) d.nofollow++;
-    d.pages.add(e.from);
-    domMap.set(domain, d);
-  }
-  const externalDomains = [...domMap.entries()].map(([domain, d]) => ({ domain, links: d.links, nofollow: d.nofollow, pages: d.pages.size })).sort((a, b) => b.links - a.links);
-  const round = (n) => Math.round(n * 10) / 10;
-  return {
-    summary: {
-      internal: {
-        totalLinks: internalEdges.length,
-        pages,
-        orphans,
-        brokenInternal,
-        avgInlinks: pages ? round(sumInlinks / pages) : 0,
-        avgOutlinks: pages ? round(sumOutlinks / pages) : 0,
-        distribution,
-        topByInlinks
-      },
-      external: {
-        totalLinks: externalTotal,
-        uniqueDomains: externalDomains.length,
-        topDomains: externalDomains.slice(0, 20)
-      }
-    },
-    externalDomains
-  };
-}
-function renderLinkReport(r) {
-  const s = r.summary;
-  const lines = [
-    `## Link analysis`,
-    `**Internal:** ${s.internal.totalLinks} links \xB7 ${s.internal.pages} pages \xB7 avg ${s.internal.avgInlinks} inlinks/page \xB7 ${s.internal.orphans} orphans \xB7 ${s.internal.brokenInternal} broken`,
-    `**External:** ${s.external.totalLinks} links to ${s.external.uniqueDomains} domains`,
-    ``,
-    `### Inlink distribution`,
-    `- 0 inlinks (orphans): ${s.internal.distribution.zero}`,
-    `- 1\u20132: ${s.internal.distribution.oneToTwo}`,
-    `- 3\u201310: ${s.internal.distribution.threeToTen}`,
-    `- 11+: ${s.internal.distribution.elevenPlus}`,
-    ``,
-    `### Top 20 internal pages by inlinks`,
-    `| inlinks | out (int/ext) | URL |`,
-    `|---|---|---|`,
-    ...s.internal.topByInlinks.map((p) => `| ${p.inlinks} | ${p.outlinksInternal}/${p.outlinksExternal} | ${p.url} |`),
-    ``,
-    `### Top 20 external domains by links`,
-    `| links | nofollow | from pages | domain |`,
-    `|---|---|---|---|`,
-    ...s.external.topDomains.map((d) => `| ${d.links} | ${d.nofollow} | ${d.pages} | ${d.domain} |`)
-  ];
-  return lines.join("\n");
+// src/services/site-architecture-auditor/site-audit-repository.ts
+function rowToSiteAuditJobRow(row) {
+  return SiteAuditJobRowSchema.parse({
+    id: String(row.id),
+    user_id: Number(row.user_id),
+    status: row.status != null ? String(row.status) : "pending",
+    client_domain: String(row.client_domain),
+    session_path: String(row.session_path),
+    request: String(row.request),
+    result: row.result != null ? String(row.result) : null,
+    error: row.error != null ? String(row.error) : null,
+    created_at: String(row.created_at),
+    updated_at: String(row.updated_at)
+  });
 }
-var init_seo_link_report = __esm({
-  "src/api/seo-link-report.ts"() {
+var SiteAuditRepository;
+var init_site_audit_repository = __esm({
+  "src/services/site-architecture-auditor/site-audit-repository.ts"() {
     "use strict";
+    init_db();
+    init_db();
+    SiteAuditRepository = class {
+      db;
+      constructor() {
+        this.db = getDb();
+      }
+      async createSiteAuditJob(jobId, userId, request) {
+        await this.db.execute({
+          sql: `INSERT INTO site_audit_jobs
+              (id, user_id, status, client_domain, session_path, request, created_at, updated_at)
+            VALUES (?, ?, 'pending', ?, ?, ?, datetime('now'), datetime('now'))`,
+          args: [
+            jobId,
+            userId,
+            request.clientDomain,
+            request.sessionPath,
+            JSON.stringify(request)
+          ]
+        });
+      }
+      async claimPendingSiteAuditJob() {
+        const upd = await this.db.execute(
+          `UPDATE site_audit_jobs
+         SET status = 'running', updated_at = datetime('now')
+       WHERE id = (
+         SELECT id FROM site_audit_jobs
+         WHERE status = 'pending'
+         ORDER BY created_at ASC
+         LIMIT 1
+       )
+       RETURNING *`
+        );
+        if (upd.rowsAffected === 0) return null;
+        const raw = upd.rows[0];
+        if (!raw) return null;
+        return rowToSiteAuditJobRow(raw);
+      }
+      async getSiteAuditJob(jobId) {
+        const res = await this.db.execute({
+          sql: `SELECT * FROM site_audit_jobs WHERE id = ?`,
+          args: [jobId]
+        });
+        const raw = res.rows[0];
+        if (!raw) return null;
+        return rowToSiteAuditJobRow(raw);
+      }
+      async listSiteAuditJobs(userId) {
+        const res = userId != null ? await this.db.execute({
+          sql: `SELECT * FROM site_audit_jobs WHERE user_id = ? ORDER BY created_at DESC LIMIT 50`,
+          args: [userId]
+        }) : await this.db.execute(
+          `SELECT * FROM site_audit_jobs ORDER BY created_at DESC LIMIT 50`
+        );
+        return res.rows.map((r) => rowToSiteAuditJobRow(r));
+      }
+      async updateSiteAuditJobState(jobId, status) {
+        await this.db.execute({
+          sql: `UPDATE site_audit_jobs SET status = ?, updated_at = datetime('now') WHERE id = ?`,
+          args: [status, jobId]
+        });
+      }
+      async completeSiteAuditJob(jobId, result) {
+        await this.db.execute({
+          sql: `UPDATE site_audit_jobs
+              SET status = 'complete', result = ?, updated_at = datetime('now')
+            WHERE id = ?`,
+          args: [JSON.stringify(result), jobId]
+        });
+      }
+      async failSiteAuditJob(jobId, error) {
+        await this.db.execute({
+          sql: `UPDATE site_audit_jobs
+              SET status = 'failed', error = ?, updated_at = datetime('now')
+            WHERE id = ?`,
+          args: [error, jobId]
+        });
+      }
+      async logSiteAuditPhaseComplete(jobId, phase, output) {
+        await this.db.execute({
+          sql: `INSERT OR IGNORE INTO site_audit_phase_log (job_id, phase, output_summary, completed_at)
+            VALUES (?, ?, ?, datetime('now'))`,
+          args: [jobId, phase, JSON.stringify(output)]
+        }).catch((err) => {
+          console.warn(
+            "[site-audit-repository] logSiteAuditPhaseComplete failed:",
+            err instanceof Error ? err.message : String(err)
+          );
+        });
+      }
+    };
   }
 });
 
-// src/api/seo-issues.ts
-function dupes(pages, key) {
-  const groups = /* @__PURE__ */ new Map();
-  for (const p of pages) {
-    const k = key(p);
-    if (k === null || k === void 0 || k === "") continue;
-    const ks = String(k);
-    if (!groups.has(ks)) groups.set(ks, []);
-    groups.get(ks).push(p.url);
-  }
-  return [...groups.values()].filter((g) => g.length > 1).flat();
-}
-function computeIssues(pages, metrics, precomputedBrokenLinkPages) {
-  const group = (urls) => ({ count: urls.length, urls: urls.slice(0, 200) });
-  const where = (fn) => pages.filter(fn).map((p) => p.url);
-  const ok = (p) => p.status === 200;
-  const pathname = (u) => {
-    try {
-      return new URL(u).pathname;
-    } catch {
-      return "";
-    }
-  };
-  const report = {
-    "title.missing": group(where((p) => ok(p) && !p.title)),
-    "title.duplicate": group(dupes(pages.filter(ok), (p) => p.title)),
-    "title.tooLong": group(where((p) => (p.titleLength ?? 0) > TITLE_MAX || (p.titlePixels ?? 0) > TITLE_PX_MAX)),
-    "title.tooShort": group(where((p) => p.title != null && (p.titleLength ?? 0) < TITLE_MIN)),
-    "title.sameAsH1": group(where((p) => !!p.title && !!p.h1 && p.title.trim() === p.h1.trim())),
-    "meta.missing": group(where((p) => ok(p) && !p.metaDescription)),
-    "meta.duplicate": group(dupes(pages.filter(ok), (p) => p.metaDescription)),
-    "meta.tooLong": group(where((p) => (p.metaDescLength ?? 0) > META_MAX)),
-    "meta.tooShort": group(where((p) => p.metaDescription != null && (p.metaDescLength ?? 0) < META_MIN)),
-    "h1.missing": group(where((p) => ok(p) && !p.h1)),
-    "h1.multiple": group(where((p) => !!p.h1_2)),
-    "h1.duplicate": group(dupes(pages.filter(ok), (p) => p.h1)),
-    "h1.tooLong": group(where((p) => (p.h1?.length ?? 0) > H1_MAX)),
-    "h2.missing": group(where((p) => ok(p) && p.h2Count === 0)),
-    "indexability.nonIndexable": group(where((p) => !p.indexable)),
-    "indexability.noindex": group(where((p) => p.indexabilityReason === "noindex")),
-    "canonical.missing": group(where((p) => ok(p) && !p.canonicalUrl)),
-    "canonical.canonicalised": group(where((p) => p.indexabilityReason === "canonicalised")),
-    "response.broken4xx": group(where((p) => (p.status ?? 0) >= 400 && (p.status ?? 0) < 500)),
-    "response.error5xx": group(where((p) => (p.status ?? 0) >= 500)),
-    "response.redirect3xx": group(where((p) => (p.status ?? 0) >= 300 && (p.status ?? 0) < 400)),
-    "response.noResponse": group(where((p) => p.status === null)),
-    "content.thin": group(where((p) => ok(p) && p.wordCount > 0 && p.wordCount < THIN_WORDS)),
-    "content.exactDuplicate": group(dupes(pages.filter((p) => ok(p) && p.contentHash), (p) => p.contentHash)),
-    "images.missingAlt": group(where((p) => p.imagesMissingAlt > 0)),
-    "schema.missing": group(where((p) => ok(p) && p.schemaTypes.length === 0)),
-    "url.tooLong": group(where((p) => p.url.length > URL_MAX)),
-    "url.uppercase": group(where((p) => /[A-Z]/.test(pathname(p.url)))),
-    "url.underscores": group(where((p) => pathname(p.url).includes("_"))),
-    "links.orphan": group([...metrics.values()].filter((m) => m.orphan).map((m) => m.url))
-  };
-  let brokenLinkPages = precomputedBrokenLinkPages;
-  if (!brokenLinkPages) {
-    const normUrl2 = (u) => {
-      try {
-        const x = new URL(u);
-        x.hash = "";
-        return (x.origin + x.pathname).replace(/\/+$/, "") + x.search;
-      } catch {
-        return u.replace(/#.*$/, "").replace(/\/+$/, "");
+// src/services/site-architecture-auditor/python/graph-runner.ts
+var childProcess, path, util, import_node_url, import_meta, execFile2, __dirname, SiteAuditPythonRunner;
+var init_graph_runner = __esm({
+  "src/services/site-architecture-auditor/python/graph-runner.ts"() {
+    "use strict";
+    childProcess = __toESM(require("child_process"), 1);
+    path = __toESM(require("path"), 1);
+    util = __toESM(require("util"), 1);
+    import_node_url = require("url");
+    init_schemas();
+    import_meta = {};
+    execFile2 = util.promisify(childProcess.execFile);
+    __dirname = path.dirname((0, import_node_url.fileURLToPath)(import_meta.url));
+    SiteAuditPythonRunner = class {
+      graphMetricsScript;
+      locationClassifierScript;
+      constructor() {
+        try {
+          childProcess.execFileSync("python3", ["-c", "import networkx"]);
+        } catch {
+          throw new Error("SiteAuditPythonRunner: networkx not available");
+        }
+        this.graphMetricsScript = path.join(__dirname, "scripts", "compute_graph_metrics.py");
+        this.locationClassifierScript = path.join(__dirname, "scripts", "location_classifier.py");
       }
-    };
-    const statusByUrl = new Map(pages.map((p) => [normUrl2(p.url), p.status]));
-    brokenLinkPages = /* @__PURE__ */ new Set();
-    for (const p of pages) {
-      for (const l of p.outlinks ?? []) {
-        if (!l.internal) continue;
-        const st = statusByUrl.get(normUrl2(l.href));
-        if (st != null && st >= 400) brokenLinkPages.add(p.url);
+      async runGraphMetrics(input) {
+        const validatedInput = PythonGraphInputSchema.parse(input);
+        let stdout;
+        let stderr;
+        try {
+          ({ stdout, stderr } = await execFile2(
+            "python3",
+            [this.graphMetricsScript, "--input", JSON.stringify(validatedInput)],
+            { timeout: 12e4 }
+          ));
+        } catch (err) {
+          const execError = err;
+          throw new Error(`python-subprocess-001: ${execError.stderr ?? String(err)}`);
+        }
+        void stderr;
+        return PythonGraphOutputSchema.parse(JSON.parse(stdout));
       }
-    }
-  }
-  report["links.brokenInternal"] = group([...brokenLinkPages]);
-  return report;
-}
-function renderIssueReport(siteUrl, pages, report, metrics) {
-  const total = pages.length;
-  const sev = (key) => key.startsWith("response.broken") || key.startsWith("response.error") || key.startsWith("links.broken") ? "\u{1F534}" : key.startsWith("title.missing") || key.startsWith("h1.missing") || key.startsWith("indexability") || key.startsWith("canonical.missing") ? "\u{1F7E0}" : "\u{1F7E1}";
-  const rows = Object.entries(report).filter(([, g]) => g.count > 0).sort((a, b) => b[1].count - a[1].count).map(([k, g]) => `| ${sev(k)} | \`${k}\` | ${g.count} | ${g.urls.slice(0, 3).join(" \xB7 ")}${g.urls.length > 3 ? " \u2026" : ""} |`);
-  const depths = [...metrics.values()].map((m) => m.crawlDepth).filter((d) => d != null);
-  const maxDepth = depths.length ? Math.max(...depths) : 0;
-  const orphans = [...metrics.values()].filter((m) => m.orphan).length;
-  const topLinked = [...metrics.values()].sort((a, b) => b.inlinks - a.inlinks).slice(0, 5);
-  return [
-    `# SEO Crawl Report: ${siteUrl}`,
-    `**${total} pages crawled** \xB7 max crawl depth ${maxDepth} \xB7 ${orphans} orphan page(s)`,
-    `
-## Issues
-| | Issue | Count | Examples |
-|---|-------|-------|----------|
-${rows.join("\n") || "| \u2705 | none | 0 | \u2014 |"}`,
-    `
-## Most-linked pages
-${topLinked.map((m) => `- ${m.inlinks} inlinks \xB7 depth ${m.crawlDepth ?? "\u2014"} \xB7 ${m.url}`).join("\n")}`,
-    `
-_Thresholds: title ${TITLE_MAX}ch/${TITLE_PX_MAX}px, meta ${META_MAX}ch, H1 ${H1_MAX}ch, thin <${THIN_WORDS} words, URL ${URL_MAX}ch. Pixel widths are estimates._`
-  ].join("\n");
+      async runLocationClassifier(input) {
+        const validatedInput = LocationClassifierInputSchema.parse(input);
+        let stdout;
+        let stderr;
+        try {
+          ({ stdout, stderr } = await execFile2(
+            "python3",
+            [this.locationClassifierScript, "--input", JSON.stringify(validatedInput)],
+            { timeout: 12e4 }
+          ));
+        } catch (err) {
+          const execError = err;
+          throw new Error(`python-subprocess-001: ${execError.stderr ?? String(err)}`);
+        }
+        void stderr;
+        return LocationClassifierOutputSchema.parse(JSON.parse(stdout));
+      }
+    };
+  }
+});
+
+// src/services/site-architecture-auditor/factory.ts
+function makeSiteAuditService() {
+  const deepInfraKey = process.env["DEEPINFRA_API_KEY"];
+  if (!deepInfraKey) throw new Error("DEEPINFRA_API_KEY is required");
+  const primary = new DeepInfraLlmClient(deepInfraKey);
+  const llm = process.env["OPENROUTER_API_KEY"] ? new FallbackLlmClient(primary, new OpenRouterLlmClient(process.env["OPENROUTER_API_KEY"])) : primary;
+  const repo = new SiteAuditRepository();
+  const http = new FetchHttpClient();
+  const python = new SiteAuditPythonRunner();
+  return new SiteAuditService({ repo, llm, http, python });
 }
-var THIN_WORDS, TITLE_MAX, TITLE_PX_MAX, TITLE_MIN, META_MAX, META_MIN, H1_MAX, URL_MAX;
-var init_seo_issues = __esm({
-  "src/api/seo-issues.ts"() {
+var init_factory = __esm({
+  "src/services/site-architecture-auditor/factory.ts"() {
     "use strict";
-    THIN_WORDS = 200;
-    TITLE_MAX = 60;
-    TITLE_PX_MAX = 561;
-    TITLE_MIN = 30;
-    META_MAX = 155;
-    META_MIN = 70;
-    H1_MAX = 70;
-    URL_MAX = 115;
+    init_llm_parse_with_retry();
+    init_http_client();
+    init_site_audit_service();
+    init_site_audit_repository();
+    init_graph_runner();
   }
 });
 
-// src/api/image-audit.ts
-function formatBytes(n) {
-  if (n == null) return null;
-  if (n < 1024) return `${n} B`;
-  if (n < 1024 * 1024) return `${(n / 1024).toFixed(1)} KB`;
-  return `${(n / (1024 * 1024)).toFixed(2)} MB`;
-}
-function collectUrls(pages) {
-  const byKey = /* @__PURE__ */ new Map();
-  for (const p of pages) for (const raw of p.imageLinks || []) {
-    const u = decodeEntities(raw);
-    if (!/^https?:\/\//i.test(u)) continue;
-    if (/[{}]|%7[bd]/i.test(u)) continue;
-    const k = dedupKey(u);
-    const prev = byKey.get(k);
-    if (!prev || /^https:/i.test(u) && /^http:/i.test(prev)) byKey.set(k, u);
+// src/services/site-architecture-auditor/phases.ts
+async function dispatchSiteAuditPhase(service, phase, raw) {
+  switch (phase) {
+    case "phase1-ingest":
+      return handleIngestPhase(service, raw);
+    case "phase2-build-graph":
+      return handleBuildGraphPhase(service, raw);
+    case "phase3-classify":
+      return handleClassifyPhase(service, raw);
+    case "phase4-compare":
+      return handleComparePhase(service, raw);
+    case "phase5-synthesize":
+      return handleSynthesizePhase(service, raw);
+    default: {
+      const _exhaustive = phase;
+      throw new Error("Unknown phase: " + String(_exhaustive));
+    }
   }
-  return [...byKey.values()];
 }
-async function sizeAndType(url, timeoutMs) {
-  const ctrl = new AbortController();
-  const t = setTimeout(() => ctrl.abort(), timeoutMs);
-  const read = (res) => ({
-    len: res.headers.get("content-length"),
-    cr: res.headers.get("content-range"),
-    ct: res.headers.get("content-type"),
-    status: res.status
-  });
-  try {
-    let bytes = null;
-    let ct = null;
-    let status = null;
-    try {
-      const h = read(await fetch(url, { method: "HEAD", redirect: "follow", signal: ctrl.signal }));
-      status = h.status;
-      ct = h.ct;
-      if (h.len) bytes = Number(h.len);
-    } catch {
-    }
-    if (bytes == null) {
-      const g = await fetch(url, { method: "GET", headers: { Range: "bytes=0-0" }, redirect: "follow", signal: ctrl.signal });
-      const r = read(g);
-      status = status ?? r.status;
-      ct = ct ?? r.ct;
-      if (r.cr && r.cr.includes("/")) {
-        const tail = r.cr.split("/")[1];
-        if (tail && tail !== "*") bytes = Number(tail);
-      } else if (r.len) bytes = Number(r.len);
-      try {
-        await g.body?.cancel();
-      } catch {
-      }
+async function handleIngestPhase(service, raw) {
+  const payload = SiteAuditIngestPayloadSchema.parse(raw);
+  await service.runIngestPhase(payload.jobId, payload);
+  await service.runEnrichPhase(payload.jobId, payload);
+}
+async function handleBuildGraphPhase(service, raw) {
+  const payload = SiteAuditBuildGraphPayloadSchema.parse(raw);
+  const limit = (0, import_p_limit2.default)(3);
+  const tasks = [
+    limit(() => service.runBuildGraphPhase(payload.jobId, payload))
+  ];
+  const results = await Promise.allSettled(tasks);
+  for (const result of results) {
+    if (result.status === "rejected") {
+      console.warn(
+        "[phases] handleBuildGraphPhase: phase failed for job",
+        payload.jobId,
+        result.reason instanceof Error ? result.reason.message : String(result.reason)
+      );
     }
-    return { url, status, bytes: Number.isFinite(bytes) ? bytes : null, contentType: ct };
-  } catch (e) {
-    return { url, status: null, bytes: null, contentType: null, error: String(e.message || e) };
-  } finally {
-    clearTimeout(t);
   }
 }
-async function pool(items, n, fn) {
-  const out = new Array(items.length);
-  let i = 0;
-  await Promise.all(Array.from({ length: Math.min(n, items.length) }, async () => {
-    while (i < items.length) {
-      const idx = i++;
-      out[idx] = await fn(items[idx]);
+async function handleClassifyPhase(service, raw) {
+  const payload = SiteAuditClassifyPayloadSchema.parse(raw);
+  const limit = (0, import_p_limit2.default)(3);
+  const tasks = [
+    limit(() => service.runClassifyPhase(payload.jobId, payload))
+  ];
+  const results = await Promise.allSettled(tasks);
+  for (const result of results) {
+    if (result.status === "rejected") {
+      console.warn(
+        "[phases] handleClassifyPhase: phase failed for job",
+        payload.jobId,
+        result.reason instanceof Error ? result.reason.message : String(result.reason)
+      );
     }
-  }));
-  return out;
+  }
 }
-async function auditImages(pages, opts = {}) {
-  const concurrency = opts.concurrency ?? 12;
-  const timeoutMs = opts.timeoutMs ?? 12e3;
-  const urls = collectUrls(pages).slice(0, opts.max ?? 5e3);
-  const heads = await pool(urls, concurrency, (u) => sizeAndType(u, timeoutMs));
-  const rows = heads.map((r) => {
-    const format = formatOf(r.contentType, r.url);
-    return {
-      ...r,
-      size: formatBytes(r.bytes),
-      format,
-      over100kb: r.bytes != null && r.bytes > OVER_BYTES,
-      legacyFormat: format !== "unknown" && !MODERN.has(format)
-    };
-  });
-  const sized = rows.filter((r) => r.bytes != null);
-  const totalBytes = sized.reduce((a, r) => a + r.bytes, 0);
-  const formatCounts = {};
-  for (const r of rows) formatCounts[r.format] = (formatCounts[r.format] || 0) + 1;
-  return {
-    rows: rows.sort((a, b) => (b.bytes ?? 0) - (a.bytes ?? 0)),
-    summary: {
-      unique: rows.length,
-      sized: sized.length,
-      totalBytes,
-      totalSize: formatBytes(totalBytes) ?? "0 B",
-      avgSize: formatBytes(Math.round(totalBytes / (sized.length || 1))) ?? "0 B",
-      over100kb: rows.filter((r) => r.over100kb).length,
-      legacyFormat: rows.filter((r) => r.legacyFormat).length,
-      formatCounts
-    }
+async function handleComparePhase(service, raw) {
+  const payload = SiteAuditComparePayloadSchema.parse(raw);
+  const compareResult = await service.runComparePhase(payload.jobId, payload);
+  const rawRecommendations = compareResult.linkRecommendations ?? [];
+  const resolveFlagsStub = (_item, _candidates) => {
+    return [];
   };
+  const guardrailResult = applyGuardrailRules(rawRecommendations, { resolveFlags: resolveFlagsStub });
+  console.warn(
+    "[phases] handleComparePhase: guardrail check complete \u2014 passed:",
+    guardrailResult.passed.length,
+    "flagged:",
+    guardrailResult.flagged.length,
+    "(context-dependent flags require site-graph state not yet exposed by runComparePhase)"
+  );
 }
-function renderImageSection(audit) {
-  const s = audit.summary;
-  const fmt = Object.entries(s.formatCounts).sort((a, b) => b[1] - a[1]).map(([k, v]) => `${k}:${v}`).join("  ");
-  const heaviest = audit.rows.filter((r) => r.over100kb).slice(0, 15);
-  const lines = [
-    `## Images`,
-    `**${s.unique} unique images** \xB7 ${s.totalSize} total \xB7 ${s.avgSize} avg \xB7 ${s.over100kb} over 100 KB \xB7 ${s.legacyFormat} legacy format`,
-    `Formats: ${fmt}`
-  ];
-  if (heaviest.length) {
-    lines.push("", "| Size | Format | URL |", "|------|--------|-----|");
-    for (const r of heaviest) lines.push(`| ${r.size} | ${r.format} | ${r.url} |`);
-  }
-  return lines.join("\n");
+async function handleSynthesizePhase(service, raw) {
+  const payload = SiteAuditSynthesizePayloadSchema.parse(raw);
+  await service.runSynthesizePhase(payload.jobId, payload);
 }
-var OVER_BYTES, MODERN, decodeEntities, dedupKey, formatOf;
-var init_image_audit = __esm({
-  "src/api/image-audit.ts"() {
+var import_p_limit2;
+var init_phases = __esm({
+  "src/services/site-architecture-auditor/phases.ts"() {
     "use strict";
-    OVER_BYTES = 100 * 1024;
-    MODERN = /* @__PURE__ */ new Set(["webp", "avif", "svg", "svg+xml"]);
-    decodeEntities = (u) => u.replace(/&amp;/g, "&").replace(/&#0?38;/g, "&").replace(/&#x26;/gi, "&").trim();
-    dedupKey = (u) => u.replace(/^https?:\/\//i, "//").replace(/\/$/, "");
-    formatOf = (ct, url) => {
-      if (ct) {
-        const m = ct.split(";")[0].trim().toLowerCase();
-        if (m.startsWith("image/")) return m.replace("image/", "");
+    import_p_limit2 = __toESM(require("p-limit"), 1);
+    init_schemas();
+    init_rules();
+  }
+});
+
+// src/inngest/functions/site-audit.ts
+var siteAuditFn;
+var init_site_audit = __esm({
+  "src/inngest/functions/site-audit.ts"() {
+    "use strict";
+    init_client();
+    init_factory();
+    init_phases();
+    siteAuditFn = inngest.createFunction(
+      { id: "site-audit", triggers: [{ event: "site-architecture-auditor/audit.requested" }] },
+      async ({ event, step }) => {
+        const { phase, payload } = event.data;
+        await step.run("dispatch-phase", async () => {
+          const service = makeSiteAuditService();
+          await dispatchSiteAuditPhase(service, phase, payload);
+        });
+        return { phase, status: "done" };
       }
-      return (url.split("?")[0].match(/\.([a-z0-9]+)$/i)?.[1] || "unknown").toLowerCase();
-    };
+    );
   }
 });
 
@@ -10912,6 +10955,7 @@ ${renderImageSection(extras.imageAudit)}`;
       (0, import_node_fs3.writeFileSync)((0, import_node_path3.join)(dir, "images.jsonl"), extras.imageAudit.rows.map((r) => JSON.stringify(r)).join("\n"));
       (0, import_node_fs3.writeFileSync)((0, import_node_path3.join)(dir, "images-summary.json"), JSON.stringify(extras.imageAudit.summary, null, 2));
     }
+    if (extras.seoAudit) (0, import_node_fs3.writeFileSync)((0, import_node_path3.join)(dir, "seo-audit.json"), JSON.stringify(extras.seoAudit, null, 2));
     if (extras.branding) (0, import_node_fs3.writeFileSync)((0, import_node_path3.join)(dir, "branding.json"), JSON.stringify(extras.branding, null, 2));
     const artifactFiles = [
       { name: "report.md", type: "text/markdown" },
@@ -10927,6 +10971,7 @@ ${renderImageSection(extras.imageAudit)}`;
       artifactFiles.push({ name: "images.jsonl", type: "application/x-ndjson" });
       artifactFiles.push({ name: "images-summary.json", type: "application/json" });
     }
+    if (extras.seoAudit) artifactFiles.push({ name: "seo-audit.json", type: "application/json" });
     if (extras.branding) artifactFiles.push({ name: "branding.json", type: "application/json" });
     const zip = new import_yazl.ZipFile();
     const zipOut = (0, import_node_fs3.createWriteStream)((0, import_node_path3.join)(dir, "bundle.zip"));
@@ -10974,6 +11019,7 @@ var init_site_extract = __esm({
     "use strict";
     init_client();
     init_site_extractor();
+    init_seo_audit();
     init_seo_link_graph();
     init_seo_link_report();
     init_seo_issues();
@@ -11012,12 +11058,13 @@ var init_site_extract = __esm({
         const maxPages = Number(job.options.maxPages ?? 1e4);
         const concurrency = Number(job.options.concurrency ?? 1);
         const urlsPerBrowser = Number(job.options.urlsPerBrowser ?? job.options.rotateProxyEvery ?? 10);
-        const seed = await step.run("discover", async () => {
-          const discovered = await discoverUrls(job.startUrl, maxPages, key);
-          const out = discovered.length ? discovered : [job.startUrl];
-          await setExtractJobTotal(jobId, out.length);
-          return out;
+        const seedRaw = await step.run("discover", async () => {
+          const discovered = await discoverUrlsWithSitemap(job.startUrl, maxPages, key);
+          const urls = discovered.urls.length ? discovered.urls : [job.startUrl];
+          await setExtractJobTotal(jobId, urls.length);
+          return { urls, sitemapUrls: discovered.sitemapUrls };
         });
+        const seed = Array.isArray(seedRaw) ? { urls: seedRaw, sitemapUrls: [] } : seedRaw;
         const siteHost = (() => {
           try {
             return new URL(job.startUrl).hostname.replace(/^www\./, "");
@@ -11041,8 +11088,9 @@ var init_site_extract = __esm({
             return false;
           }
         };
-        const queue = [...seed];
-        const seen = new Set(seed.map(norm));
+        const sitemapSet = seed.sitemapUrls.length > 0 ? new Set(seed.sitemapUrls.map(norm)) : null;
+        const queue = [...seed.urls];
+        const seen = new Set(seed.urls.map(norm));
         let crawled = 0;
         const MAX_BATCHES = Math.ceil(maxPages / BATCH) + 30;
         for (let i = 0; i < MAX_BATCHES && crawled < maxPages && crawled < queue.length; i++) {
@@ -11050,7 +11098,7 @@ var init_site_extract = __esm({
           if (!batch.length) break;
           const newLinks = await step.run(`crawl-batch-${i}`, async () => {
             const pages = await extractPagesRotating(batch, { kernelApiKey: key, concurrency, urlsPerBrowser });
-            await saveExtractPages(jobId, pages);
+            await saveExtractPages(jobId, pages.map((p) => ({ ...p, inSitemap: sitemapSet ? sitemapSet.has(norm(p.url)) : null })));
             const found = [];
             for (const p of pages) for (const l of p.outlinks ?? []) {
               if (l.internal && l.href && sameSite(l.href)) found.push(l.href);
@@ -11074,9 +11122,13 @@ var init_site_extract = __esm({
           const pages = await getExtractedPages(jobId);
           return auditImages(pages, { concurrency: 12, timeoutMs: 12e3 });
         }) : null;
+        const seoAudit = Array.isArray(job.options.formats) && job.options.formats.includes("issues") ? await step.run("seo-audit", async () => {
+          const pages = await getExtractedPages(jobId);
+          return buildSeoAudit(pages, job.startUrl);
+        }) : null;
         const artifacts = await step.run("finalize", async () => {
           const { assembleExtractArtifacts: assembleExtractArtifacts2 } = await Promise.resolve().then(() => (init_extract_bundle(), extract_bundle_exports));
-          const stored = await assembleExtractArtifacts2(job, { branding, imageAudit });
+          const stored = await assembleExtractArtifacts2(job, { branding, imageAudit, seoAudit });
           await completeExtractJob(jobId, stored);
           return stored;
         });
@@ -13963,7 +14015,7 @@ var init_server_schemas = __esm({
       rotateProxies: import_zod13.z.boolean().optional(),
       rotateProxyEvery: import_zod13.z.number().int().min(1).max(100).optional(),
       background: import_zod13.z.boolean().optional(),
-      formats: import_zod13.z.array(import_zod13.z.enum(["markdown", "links", "json", "images", "branding"])).optional()
+      formats: import_zod13.z.array(import_zod13.z.enum(["markdown", "links", "json", "images", "branding", "issues"])).optional()
     });
     YoutubeHarvestBodySchema = import_zod13.z.object({
       mode: import_zod13.z.enum(["search", "channel"]),
@@ -27400,7 +27452,7 @@ var PACKAGE_VERSION;
 var init_version = __esm({
   "src/version.ts"() {
     "use strict";
-    PACKAGE_VERSION = "0.7.0";
+    PACKAGE_VERSION = "0.8.0";
   }
 });
 
@@ -38839,6 +38891,7 @@ async function createSignupStripeCustomer(email) {
 }
 function opForCostPath(p) {
   if (p === "/harvest" || p === "/harvest/sync") return "harvest";
+  if (p === "/diff-page") return "diff_page";
   if (p === "/extract-url") return "page_scrape";
   if (p === "/extract-site") return "extract_site";
   if (p === "/map-urls") return "url_map";
@@ -39182,6 +39235,13 @@ var init_server = __esm({
       c.header("Cache-Control", "public, max-age=60");
       return c.json(CATALOG);
     });
+    app.get("/rates", (c) => {
+      const costs = CREDIT_COST_CATALOG.map(({ aliases, ...cost }) => ({
+        ...cost,
+        ...cost.notes ? { notes: cost.notes.replace(/ via fal\.ai/g, "") } : {}
+      }));
+      return c.json({ creditsPerDollarStarter: Math.round(1 / 15e-5), costs });
+    });
     app.get("/me", async (c) => {
       const token = (0, import_cookie2.getCookie)(c, "session");
       if (!token) return c.json({ authenticated: false });