@canonry/canonry 4.161.0 → 4.163.0

package/dist/{chunk-J27J2DF6.js → chunk-G5UZTODF.js}

@@ -183,6 +183,7 @@ import {
   buildHarvestAnchorTerms,
   buildMeasurementExecutionIdentity,
   buildMeasurementRunManifestV1,
+  calendarDateRange,
   calendarMonthBounds,
   canonicalMeasurementExecutionIdentityJson,
   canonicalMeasurementPlanJson,
@@ -324,6 +325,7 @@ import {
   keywordDtoSchema,
   keywordGenerateRequestSchema,
   latestProjectRunDtoSchema,
+  linearTrend,
   locationContextSchema,
   loginRequestSchema,
   mapWithConcurrency,
@@ -537,8 +539,9 @@ import {
   surfaceClassFromCompetitorType,
   surfaceClassLabel,
   telemetryEventAcceptedDtoSchema,
-  templateLinkDetection,
   templateLinkRatio,
+  templateLinkSource,
+  templateLinkUbiquityAvailable,
   textContainsAnyBrandAlias,
   textContainsBrandAlias,
   textContainsDomain,
@@ -578,7 +581,7 @@ import {
   wordpressSchemaDeployResultDtoSchema,
   wordpressSchemaStatusResultDtoSchema,
   wordpressStatusDtoSchema
-} from "./chunk-EXDO7EPL.js";
+} from "./chunk-ESH3LPYQ.js";
 
 // src/intelligence-service.ts
 import { eq as eq59, desc as desc26, asc as asc11, and as and47, ne as ne7, or as or13, inArray as inArray21, gte as gte14, lte as lte12 } from "drizzle-orm";
@@ -1342,6 +1345,14 @@ var siteCrawlSnapshots = sqliteTable("site_crawl_snapshots", {
    * pass for "this site has no nav".
    */
   templateDetection: text("template_detection"),
+  /**
+   * The crawler's landmark ruleset version, from the crawl summary's
+   * `linkPlacementRulesetVersion`. NULL means this scan recorded no placement
+   * at all, which is the one thing that decides whether its links could be
+   * classified by where they sit or only by how often they repeat. It cannot be
+   * backfilled: a pre-4.7.0 crawl never observed placement.
+   */
+  linkPlacementRulesetVersion: text("link_placement_ruleset_version"),
   createdAt: text("created_at").notNull(),
   updatedAt: text("updated_at").notNull()
 }, (table) => [
@@ -1435,28 +1446,46 @@ var siteCrawlEdges = sqliteTable("site_crawl_edges", {
   nofollowOccurrences: integer("nofollow_occurrences").notNull().default(0),
   anchors: text("anchors", { mode: "json" }).$type().notNull().default([]),
   /**
-   * True when EVERY (target page, anchor) pair on this link appears on at
-   * least `TEMPLATE_LINK_RATIO_THRESHOLD` of the attempt's fetched pages: a
-   * nav, header, or footer link rather than an editorial one. Computed once at
-   * publish time by the contract's `classifyTemplateLinks`, so the map, the
-   * API filters, and the agents all read the same decision.
+   * True for a nav, header, or footer link rather than an editorial one.
+   * Computed once at publish time by the contract's `classifyTemplateLinkEdge`,
+   * so the map, the API filters, and the agents all read the same decision.
    *
-   * One row aggregates every anchor between the same two pages, so a link that
-   * is BOTH a footer link and an in-prose link is one row with two anchors. It
-   * is editorial, because the in-prose link exists.
+   * The rule is DOM placement where the crawl recorded it: any occurrence
+   * inside a main or article landmark makes the whole link editorial, and
+   * navigation with no content occurrence makes it chrome. Where the page
+   * declares no landmark that answers the question, `template_ratio` below is
+   * the fallback.
    *
-   * NULL means this row was never classified, which reads report as
-   * `unavailable-legacy-scan`. It never means "not a template link": the
-   * too-few-pages guard writes an explicit `false` and records its reason on
-   * the snapshot instead.
+   * Classification writes a STRICT BOOLEAN and never a NULL. A link no rule
+   * could measure is a real `false` whose `templateSource` reads `unmeasured`,
+   * because "not shown to be chrome" is what a content link means here. A third
+   * stored state would be invisible to every reader that treats this as a
+   * boolean (the layout input, the graph sample, the totals, the map legend,
+   * the inspector tiles) while only the two SQL link filters honoured it, and
+   * one definition of a content link is the whole point of the column.
+   *
+   * NULL therefore means exactly one thing: this row predates classification
+   * entirely, which reads report as `unavailable-legacy-scan`. It never means
+   * "not a template link".
    */
   isTemplate: integer("is_template", { mode: "boolean" }),
   /**
-   * Share of fetched pages carrying this link's LEAST ubiquitous anchor, which
-   * is the one that decides `is_template`. NULL when the link has no
-   * measurable pair (an unresolved target or no anchor).
+   * Share of fetched pages carrying this link's LEAST ubiquitous anchor. NULL
+   * when the link has no measurable pair (an unresolved target or no anchor)
+   * OR when DOM placement decided it, because it is the fallback rule's own
+   * evidence and had no vote.
    */
   templateRatio: real("template_ratio"),
+  /**
+   * Occurrences of this link split by where they sat in the source page, from
+   * the crawler's `placementOccurrences`. NULL means this scan recorded no
+   * placement for this link, which is a real state (a pre-4.7.0 crawl) and
+   * never zeros: `{0, 0, 0}` is what a redirect or canonical edge legitimately
+   * carries, since a non-anchor edge has no position in a page.
+   */
+  placementNavigationOccurrences: integer("placement_navigation_occurrences"),
+  placementContentOccurrences: integer("placement_content_occurrences"),
+  placementUnknownOccurrences: integer("placement_unknown_occurrences"),
   createdAt: text("created_at").notNull(),
   updatedAt: text("updated_at").notNull()
 }, (table) => [
@@ -2999,7 +3028,7 @@ function* streamSiteCrawlTemplateLinkEdges(tx, attemptId, batchSize = 2e3) {
   let after = "";
   for (; ; ) {
     const rows = tx.all(sql2`
-      SELECT edge_key, source_node_key, target_node_key, anchors
+      SELECT edge_key, source_node_key, target_node_key, anchors, relation
       FROM site_crawl_edges
       WHERE attempt_id = ${attemptId} AND edge_key > ${after}
       ORDER BY edge_key
@@ -3010,7 +3039,8 @@ function* streamSiteCrawlTemplateLinkEdges(tx, attemptId, batchSize = 2e3) {
       edgeKey: row.edge_key,
       sourceNodeKey: row.source_node_key,
       targetNodeKey: row.target_node_key,
-      anchors: parseJsonColumn(row.anchors, [])
+      anchors: parseJsonColumn(row.anchors, []),
+      relation: row.relation
     }));
     after = rows[rows.length - 1].edge_key;
     if (rows.length < batchSize) return;
@@ -3033,7 +3063,8 @@ function backfillSiteCrawlTemplateLinks(tx) {
       ON s.project_id = a.project_id AND s.run_id = a.run_id AND s.attempt_id = a.id
   `));
   for (const attempt of attempts) {
-    const detection = templateLinkDetection(attempt.pages_fetched);
+    const ubiquityAvailable = templateLinkUbiquityAvailable(attempt.pages_fetched);
+    const detection = ubiquityAvailable ? SiteHealthTemplateDetections.applied : SiteHealthTemplateDetections["unavailable-too-few-pages"];
     const scope = { attemptId: attempt.attempt_id, projectId: attempt.project_id, runId: attempt.run_id };
     tx.run(sql2`
       UPDATE site_crawl_edges SET is_template = 0, template_ratio = NULL
@@ -6204,6 +6235,35 @@ var MIGRATION_VERSIONS = [
         updated_at          TEXT NOT NULL
       )`
     ]
+  },
+  {
+    version: 138,
+    name: "site-crawl-link-placement",
+    // The crawler (aeo-audit 4.7.0) now reports where each link occurrence sat
+    // in its page, from the page's own landmarks. That is ground truth about a
+    // link, and it replaces ubiquity as the rule that separates nav from
+    // content: ubiquity keys on (target URL, anchor text), so it cannot see an
+    // editorial link whose anchor text matches the nav's, which is the common
+    // case because good anchor text reuses the destination's name. Measured on
+    // canonry.ai: 53 newly added editorial links moved the content-link count
+    // by ZERO.
+    //
+    // These columns are deliberately nullable and deliberately NOT backfilled.
+    // A crawl captured before the landmark ruleset existed never observed
+    // placement, so there is nothing to derive from and any value written here
+    // would be invented. Those scans keep the ubiquity classification they
+    // already have, and their snapshots keep a NULL ruleset version, which is
+    // exactly what makes a read say `applied` rather than `applied-placement`.
+    // Re-running a scan is what upgrades it.
+    //
+    // Idempotent: `ALTER TABLE ADD COLUMN` errors with "duplicate column name",
+    // which the runner already swallows, and there is no data pass to repeat.
+    statements: [
+      `ALTER TABLE site_crawl_edges ADD COLUMN placement_navigation_occurrences INTEGER`,
+      `ALTER TABLE site_crawl_edges ADD COLUMN placement_content_occurrences INTEGER`,
+      `ALTER TABLE site_crawl_edges ADD COLUMN placement_unknown_occurrences INTEGER`,
+      `ALTER TABLE site_crawl_snapshots ADD COLUMN link_placement_ruleset_version TEXT`
+    ]
   }
 ];
 function addRunsMeasurementPlanVersionForeignKey(tx) {
@@ -28443,7 +28503,7 @@ var crawlLimitParameter = {
 var linkKindParameter = {
   name: "linkKind",
   in: "query",
-  description: 'Restrict links to `content` (excludes nav, header, and footer links) or `template` (only those). Defaults to `all`. A scan whose links were never classified matches neither, which is why every link response also carries `templateDetection`: an empty `content` list means "could not tell" when detection is not `applied`.',
+  description: 'Restrict links to `content` (excludes nav, header, and footer links) or `template` (only those). Defaults to `all`. A link neither rule could classify matches neither filter, which is why every link response also carries `templateDetection` and every edge carries `templateSource`: an empty `content` list means "could not tell" under an `unavailable-*` detection, and `templateSource` says whether a link that IS classified was decided by where it sits in the page (`placement`) or by how many pages repeat it (`ubiquity`).',
   schema: { type: "string", enum: ["all", "content", "template"], default: "all" }
 };
 var measurementOverviewSortParameter = {
@@ -34008,7 +34068,7 @@ var routeCatalog = [
     method: "get",
     path: "/api/v1/projects/{name}/technical-aeo/graph",
     summary: "Get a persisted Site Health graph projection",
-    description: "Returns the deterministic graph projection computed once when the latest complete or selected historical crawl was published. ForceAtlas2 positions and the exact internal-anchor edge sample are persisted, so reads run no layout physics and never rescan the crawl edge table. Nav, header, and footer links are excluded from the layout physics but retained in the sample and tagged `isTemplate`, so a viewer can draw them without a refetch and without any node moving. The response is bounded to 20,000 nodes / 50,000 edges; `layout`, `omittedNodes`, `omittedEdges`, and `sampled` disclose legacy/unavailable layouts and intentional truncation, and `templateDetection` says whether template links could be told apart at all.",
+    description: "Returns the deterministic graph projection computed once when the latest complete or selected historical crawl was published. ForceAtlas2 positions and the exact internal-anchor edge sample are persisted, so reads run no layout physics and never rescan the crawl edge table. Nav, header, and footer links are excluded from the layout physics but retained in the sample and tagged `isTemplate`, so a viewer can draw them without a refetch and without any node moving. The response is bounded to 20,000 nodes / 50,000 edges; `layout`, `omittedNodes`, `omittedEdges`, and `sampled` disclose legacy/unavailable layouts and intentional truncation, and `templateDetection` says whether template links could be told apart at all and by which rule.",
     tags: ["technical-aeo"],
     parameters: [
       nameParameter,
@@ -34148,7 +34208,7 @@ var routeCatalog = [
     method: "get",
     path: "/api/v1/projects/{name}/technical-aeo/internal-links",
     summary: "List persisted internal crawl links",
-    description: "Cursor-paged internal edges for the latest or selected crawl. Optional source/target/followability/link-kind filters remain project-, run-, and attempt-scoped. `total` counts exactly what the requested filters match, and `templateDetection` says whether nav and footer links could be told apart for this scan.",
+    description: "Cursor-paged internal edges for the latest or selected crawl. Optional source/target/followability/link-kind filters remain project-, run-, and attempt-scoped. `total` counts exactly what the requested filters match, `templateDetection` says whether nav and footer links could be told apart for this scan and by which rule, and the `templateSource` plus `placementOccurrences` on each edge show the evidence behind its own classification.",
     tags: ["technical-aeo"],
     parameters: [
       nameParameter,
@@ -37627,6 +37687,7 @@ async function googleRoutes(app, opts) {
       clicks: sql13`COALESCE(SUM(${gscSearchData.clicks}), 0)`,
       impressions: sql13`COALESCE(SUM(${gscSearchData.impressions}), 0)`
     }).from(gscSearchData).where(and30(...conditions)).groupBy(gscSearchData.date).orderBy(gscSearchData.date).all();
+    const propertyDates = new Set(dailyTotals.map((d) => d.date));
     const daily = mergeGscDailyTotalsWithFallback(
       dailyTotals,
       dimensionedRows.map((r) => ({
@@ -37639,15 +37700,36 @@ async function googleRoutes(app, opts) {
       date: d.date,
       clicks: d.clicks,
       impressions: d.impressions,
-      ctr: d.impressions > 0 ? d.clicks / d.impressions : 0
+      ctr: d.impressions > 0 ? d.clicks / d.impressions : 0,
+      position: propertyDates.has(d.date) ? d.position : null
     }));
     const totalClicks = daily.reduce((sum, d) => sum + d.clicks, 0);
     const totalImpressions = daily.reduce((sum, d) => sum + d.impressions, 0);
+    let positionWeight = 0;
+    let positionWeighted = 0;
+    for (const d of daily) {
+      if (d.position === null || d.impressions <= 0) continue;
+      positionWeight += d.impressions;
+      positionWeighted += d.position * d.impressions;
+    }
+    const byDate = new Map(daily.map((d) => [d.date, d]));
+    const denseDates = daily.length === 0 ? [] : calendarDateRange(daily[0].date, daily[daily.length - 1].date);
+    const densify = (pick) => denseDates.map((date) => {
+      const row = byDate.get(date);
+      return row ? pick(row) : null;
+    });
     return {
       totals: {
         clicks: totalClicks,
         impressions: totalImpressions,
         ctr: totalImpressions > 0 ? totalClicks / totalImpressions : 0,
+        position: positionWeight > 0 ? positionWeighted / positionWeight : null,
+        /**
+         * How many days actually carried a property-level position. Below
+         * `days`, the position figure describes a SUBSET of the window, and a
+         * surface must say so rather than present it as the window's average.
+         */
+        positionDays: daily.filter((d) => d.position !== null).length,
         days: daily.length
       },
       daily,
@@ -37661,7 +37743,16 @@ async function googleRoutes(app, opts) {
       // contain the rows beside it. When only one side is given, the other is
       // dropped rather than reported reversed: an absent bound is honest about
       // being unspecified, a reversed pair is not.
-      window: resolveReportedWindow(resolvedWindow, startDate, endDate)
+      window: resolveReportedWindow(resolvedWindow, startDate, endDate),
+      // Fitted server-side so the dashboard, the CLI, and the report all draw
+      // the SAME line (UI/CLI parity — a chart-only regression is invisible to
+      // an agent), and fitted over the CALENDAR, not over the rows.
+      trends: {
+        clicks: linearTrend(densify((d) => d.clicks)),
+        impressions: linearTrend(densify((d) => d.impressions)),
+        ctr: linearTrend(densify((d) => d.ctr)),
+        position: linearTrend(densify((d) => d.position))
+      }
     };
   });
   app.get("/projects/:name/google/gsc/top-pages", async (request) => {
@@ -59091,7 +59182,15 @@ function countLivePageHealthChecks(auditFields) {
   }
   return checks;
 }
-function mapCrawlEdge(row) {
+function mapPlacementOccurrences(row) {
+  const navigation = row.placementNavigationOccurrences;
+  const content = row.placementContentOccurrences;
+  const unknown = row.placementUnknownOccurrences;
+  if (navigation == null || content == null || unknown == null) return null;
+  return { navigation, content, unknown };
+}
+function mapCrawlEdge(row, detection) {
+  const placementOccurrences = mapPlacementOccurrences(row);
   return {
     edgeKey: row.edgeKey,
     sourceNodeKey: row.sourceNodeKey,
@@ -59106,7 +59205,13 @@ function mapCrawlEdge(row) {
     nofollowOccurrences: row.nofollowOccurrences,
     anchors: row.anchors,
     isTemplate: row.isTemplate,
-    templateRatio: row.templateRatio
+    templateRatio: row.templateRatio,
+    templateSource: templateLinkSource(detection, {
+      isTemplate: row.isTemplate,
+      templateRatio: row.templateRatio,
+      placementOccurrences
+    }),
+    placementOccurrences
   };
 }
 function templateDetectionOf(value) {
@@ -59739,7 +59844,8 @@ async function technicalAeoRoutes(app, opts) {
       distance: distances.get(row.nodeKey) ?? 0,
       relationToFocus: row.nodeKey === focus.nodeKey ? "focus" : directRelations.get(row.nodeKey) ?? "transitive"
     })).sort((a, b) => a.distance - b.distance || a.nodeKey.localeCompare(b.nodeKey));
-    const edges = [...edgeRows.values()].filter((edge) => edge.targetNodeKey && persistedNodeKeys.has(edge.sourceNodeKey) && persistedNodeKeys.has(edge.targetNodeKey)).sort((a, b) => a.edgeKey.localeCompare(b.edgeKey)).map(mapCrawlEdge);
+    const subgraphDetection = templateDetectionOf(snapshot.templateDetection);
+    const edges = [...edgeRows.values()].filter((edge) => edge.targetNodeKey && persistedNodeKeys.has(edge.sourceNodeKey) && persistedNodeKeys.has(edge.targetNodeKey)).sort((a, b) => a.edgeKey.localeCompare(b.edgeKey)).map((row) => mapCrawlEdge(row, subgraphDetection));
     const omittedNodes = omittedNodeKeys.size + Math.max(0, distances.size - pageRows.length);
     const omittedEdges = omittedEdgeKeys.size + (queryTruncated ? 1 : 0) + Math.max(0, edgeRows.size - edges.length);
     const truncated = omittedNodes > 0 || omittedEdges > 0;
@@ -59888,6 +59994,7 @@ async function technicalAeoRoutes(app, opts) {
       inArray20(siteCrawlPages.nodeKey, pathKeys)
     )).all();
     const pageByKey = new Map(pathRows.map((row) => [row.nodeKey, row]));
+    const pathDetection = templateDetectionOf(snapshot.templateDetection);
     return {
       project: project.name,
       runId: snapshot.runId,
@@ -59899,7 +60006,7 @@ async function technicalAeoRoutes(app, opts) {
       maxDepth,
       visitedNodes: visited.size,
       nodes: pathKeys.map((key) => mapCrawlPage(pageByKey.get(key))),
-      edges: pathEdges.map(mapCrawlEdge)
+      edges: pathEdges.map((row) => mapCrawlEdge(row, pathDetection))
     };
   });
   app.get("/projects/:name/technical-aeo/changes", async (request) => {
@@ -60177,8 +60284,10 @@ async function technicalAeoRoutes(app, opts) {
     )).all();
     const beforePages = new Map(pageRowsFor(beforeScope).map((row) => [row.nodeKey, mapCrawlPage(row)]));
     const afterPages = new Map(pageRowsFor(afterScope).map((row) => [row.nodeKey, mapCrawlPage(row)]));
-    const beforeLinks = new Map(edgeRowsFor(beforeScope).map((row) => [row.edgeKey, mapCrawlEdge(row)]));
-    const afterLinks = new Map(edgeRowsFor(afterScope).map((row) => [row.edgeKey, mapCrawlEdge(row)]));
+    const beforeDetection = templateDetectionOf(beforeSnapshot.templateDetection);
+    const afterDetection = templateDetectionOf(afterSnapshot.templateDetection);
+    const beforeLinks = new Map(edgeRowsFor(beforeScope).map((row) => [row.edgeKey, mapCrawlEdge(row, beforeDetection)]));
+    const afterLinks = new Map(edgeRowsFor(afterScope).map((row) => [row.edgeKey, mapCrawlEdge(row, afterDetection)]));
     const changes = visibleKeys.map((row) => {
       if (row.entity === "page") {
         const before2 = beforePages.get(row.key) ?? null;
@@ -60469,7 +60578,7 @@ async function technicalAeoRoutes(app, opts) {
       nextCursor: nextOffset < total ? encodeCursor2(nextOffset) : null,
       templateDetection,
       linkKind,
-      edges: rows.map(mapCrawlEdge)
+      edges: rows.map((row) => mapCrawlEdge(row, templateDetection))
     };
   });
   app.get("/projects/:name/technical-aeo/internal-links/neighbors", async (request) => {
@@ -60532,8 +60641,8 @@ async function technicalAeoRoutes(app, opts) {
       url: request.query.url ?? null,
       templateDetection,
       linkKind,
-      inbound: inboundRows.slice(0, limit).map(mapCrawlEdge),
-      outbound: outboundRows.slice(0, limit).map(mapCrawlEdge),
+      inbound: inboundRows.slice(0, limit).map((row) => mapCrawlEdge(row, templateDetection)),
+      outbound: outboundRows.slice(0, limit).map((row) => mapCrawlEdge(row, templateDetection)),
       inboundTruncated: inboundRows.length > limit,
       outboundTruncated: outboundRows.length > limit
     };

package/dist/{chunk-HCGTKQCG.js → chunk-INXI6PO7.js}

@@ -81,7 +81,7 @@ import {
   trafficConnectWordpressRequestSchema,
   trafficEventKindSchema,
   trafficSeriesGranularitySchema
-} from "./chunk-EXDO7EPL.js";
+} from "./chunk-ESH3LPYQ.js";
 
 // src/config.ts
 import fs from "fs";
@@ -8637,7 +8637,7 @@ var technicalAeoStructureInputSchema = z3.object({
   limit: z3.number().int().positive().max(100).optional()
 });
 var linkKindSchema = z3.enum(["all", "content", "template"]).optional().describe(
-  "Restrict to content links (excludes nav, header, and footer links) or to template links only. Defaults to all. Check templateDetection before reading an empty content list as a real zero."
+  "Restrict to content links (excludes nav, header, and footer links) or to template links only. Defaults to all. Check templateDetection before reading an empty content list as a real zero, and before comparing a count with an older scan: it says whether the split came from where each link sits in the page or from how many pages repeat it."
 );
 var technicalAeoInternalLinksInputSchema = z3.object({
   project: projectNameSchema,
@@ -10651,7 +10651,7 @@ var canonryMcpTools = [
   defineTool({
     name: "canonry_technical_aeo_internal_links",
     title: "List Technical AEO internal links",
-    description: "Read a bounded, cursor-paged list of persisted internal crawl edges. Filter by source URL, target URL, followability, or link kind. Nav, header, and footer links are marked isTemplate; templateDetection reports whether that classification ran, so an empty content-only list is not evidence of no content links. Use the neighbors tool for one page rather than loading a graph.",
+    description: "Read a bounded, cursor-paged list of persisted internal crawl edges. Filter by source URL, target URL, followability, or link kind. Nav, header, and footer links are marked isTemplate; templateSource says which rule decided each one (placement, ubiquity, or unmeasured when neither had evidence) and templateDetection reports the same for the scan, so an empty content-only list is not evidence of no content links and two scans classified by different rules are never compared as if they were the same measurement. Use the neighbors tool for one page rather than loading a graph.",
     access: "read",
     tier: "monitoring",
     inputSchema: technicalAeoInternalLinksInputSchema,

package/dist/{chunk-W6OWEPZG.js → chunk-MRHYSEZB.js}

@@ -10,7 +10,7 @@ import {
   loadConfig,
   loadConfigRaw,
   saveConfigPatch
-} from "./chunk-HCGTKQCG.js";
+} from "./chunk-INXI6PO7.js";
 import {
   CC_CACHE_DIR,
   DUCKDB_SPEC,
@@ -139,7 +139,7 @@ import {
   siteCrawlSnapshots,
   toAlertView,
   usageCounters
-} from "./chunk-J27J2DF6.js";
+} from "./chunk-G5UZTODF.js";
 import {
   AGENT_MEMORY_VALUE_MAX_BYTES,
   AGENT_PROVIDER_IDS,
@@ -172,7 +172,7 @@ import {
   SITE_CRAWL_GRAPH_MAX_NODES,
   SKILL_MANIFEST_FILENAME,
   SchedulableRunKinds,
-  SiteHealthTemplateDetections,
+  SiteHealthLinkClassificationSources,
   SkillsClients,
   TEMPLATE_LINK_RATIO_THRESHOLD,
   TrafficEventConfidences,
@@ -193,11 +193,13 @@ import {
   bucketOnboardingCount,
   buildRunErrorFromMessages,
   classifySkillFile,
+  classifyTemplateLinkEdge,
   cloudflareQueueNameSchema,
   coerceSkillManifest,
   computeBacklinkSummaryMetrics,
   contentActionLabel,
   contentBriefDtoSchema,
+  createTemplateLinkDetectionTally,
   createTemplateLinkPairIndex,
   deriveCitedUrlCandidates,
   deriveIndexCoverage,
@@ -220,7 +222,6 @@ import {
   isGhostTelemetryEvent,
   isRetryableHttpError,
   isSelfLink,
-  isTemplateLinkRatio,
   isVertexGroundingRedirect,
   mapWithConcurrency,
   normalizeIdTokens,
@@ -229,6 +230,7 @@ import {
   normalizeSiteAuditRunRequest,
   normalizeUrlPath,
   notFound,
+  observeTemplateLinkDetection,
   observeTemplateLinkEdges,
   parseMeasurementRunManifestV1,
   parseOriginList,
@@ -244,13 +246,14 @@ import {
   splitList,
   startOfDayHourInTimeZone,
   templateLinkDetection,
-  templateLinkRatio,
+  templateLinkPlacementAvailable,
+  templateLinkUbiquityAvailable,
   textContainsBrandAlias,
   textContainsDomain,
   validationError,
   winnabilityClassLabel,
   withRetry
-} from "./chunk-EXDO7EPL.js";
+} from "./chunk-ESH3LPYQ.js";
 
 // src/telemetry.ts
 import crypto from "crypto";
@@ -7304,7 +7307,7 @@ var COMPONENT_PACKING_SPACING_FACTOR = 3;
 function siteCrawlGraphLayoutSettingsFingerprint(settings) {
   return crypto16.createHash("sha256").update(JSON.stringify(settings)).digest("hex").slice(0, 8);
 }
-var SITE_CRAWL_GRAPH_LAYOUT_ALGORITHM = "site-health-fa2-v4";
+var SITE_CRAWL_GRAPH_LAYOUT_ALGORITHM = "site-health-fa2-v5";
 var SITE_CRAWL_GRAPH_LAYOUT_VERSION = `${SITE_CRAWL_GRAPH_LAYOUT_ALGORITHM}-${siteCrawlGraphLayoutSettingsFingerprint({
   forceAtlas: FORCE_ATLAS_SETTINGS,
   noverlap: NOVERLAP_SETTINGS,
@@ -7860,6 +7863,13 @@ function deleteSiteCrawlGraphLayout(db, scope) {
 import { and as and11, asc as asc2, eq as eq14, gt, inArray as inArray4 } from "drizzle-orm";
 var TEMPLATE_LINK_READ_BATCH = 2e3;
 var TEMPLATE_LINK_WRITE_BATCH = 500;
+function placementOccurrencesOf(row) {
+  const { placementNavigationOccurrences: navigation } = row;
+  const { placementContentOccurrences: content } = row;
+  const { placementUnknownOccurrences: unknown } = row;
+  if (navigation == null || content == null || unknown == null) return null;
+  return { navigation, content, unknown };
+}
 function* streamAttemptEdges(db, scope) {
   let after = "";
   for (; ; ) {
@@ -7867,7 +7877,11 @@ function* streamAttemptEdges(db, scope) {
       edgeKey: siteCrawlEdges.edgeKey,
       sourceNodeKey: siteCrawlEdges.sourceNodeKey,
       targetNodeKey: siteCrawlEdges.targetNodeKey,
-      anchors: siteCrawlEdges.anchors
+      anchors: siteCrawlEdges.anchors,
+      relation: siteCrawlEdges.relation,
+      placementNavigationOccurrences: siteCrawlEdges.placementNavigationOccurrences,
+      placementContentOccurrences: siteCrawlEdges.placementContentOccurrences,
+      placementUnknownOccurrences: siteCrawlEdges.placementUnknownOccurrences
     }).from(siteCrawlEdges).where(and11(
       eq14(siteCrawlEdges.projectId, scope.projectId),
       eq14(siteCrawlEdges.runId, scope.runId),
@@ -7875,7 +7889,14 @@ function* streamAttemptEdges(db, scope) {
       gt(siteCrawlEdges.edgeKey, after)
     )).orderBy(asc2(siteCrawlEdges.edgeKey)).limit(TEMPLATE_LINK_READ_BATCH).all();
     if (rows.length === 0) return;
-    yield rows;
+    yield rows.map((row) => ({
+      edgeKey: row.edgeKey,
+      sourceNodeKey: row.sourceNodeKey,
+      targetNodeKey: row.targetNodeKey,
+      anchors: row.anchors,
+      relation: row.relation,
+      placementOccurrences: placementOccurrencesOf(row)
+    }));
     after = rows[rows.length - 1].edgeKey;
     if (rows.length < TEMPLATE_LINK_READ_BATCH) return;
   }
@@ -7885,39 +7906,56 @@ function* chunked(values, size = TEMPLATE_LINK_WRITE_BATCH) {
     yield values.slice(offset, offset + size);
   }
 }
-function classifySiteCrawlTemplateLinks(db, scope, pagesFetched) {
+function writeGroupKey(isTemplate, templateRatio) {
+  return `${isTemplate ? "1" : "0"}${templateRatio ?? ""}`;
+}
+function classifySiteCrawlTemplateLinks(db, scope, pagesFetched, placementRulesetVersion) {
   const scopeFilter = and11(
     eq14(siteCrawlEdges.projectId, scope.projectId),
     eq14(siteCrawlEdges.runId, scope.runId),
     eq14(siteCrawlEdges.attemptId, scope.attemptId)
   );
-  const detection = templateLinkDetection(pagesFetched);
+  const placementAvailable = templateLinkPlacementAvailable(placementRulesetVersion);
+  const ubiquityAvailable = templateLinkUbiquityAvailable(pagesFetched);
   db.update(siteCrawlEdges).set({ isTemplate: false, templateRatio: null }).where(scopeFilter).run();
-  if (detection !== SiteHealthTemplateDetections.applied) {
-    return { detection, templateEdgeCount: 0 };
-  }
   const index = createTemplateLinkPairIndex();
-  for (const batch of streamAttemptEdges(db, scope)) observeTemplateLinkEdges(index, batch);
-  const edgeKeysByRatio = /* @__PURE__ */ new Map();
+  if (ubiquityAvailable) {
+    for (const batch of streamAttemptEdges(db, scope)) observeTemplateLinkEdges(index, batch);
+  }
+  const edgeKeysByValue = /* @__PURE__ */ new Map();
   let templateEdgeCount = 0;
+  let placementEdgeCount = 0;
+  const tally = createTemplateLinkDetectionTally();
   for (const batch of streamAttemptEdges(db, scope)) {
     for (const edge of batch) {
-      const ratio = templateLinkRatio(index, pagesFetched, edge);
-      if (ratio == null) continue;
-      const group = edgeKeysByRatio.get(ratio);
-      if (group) group.push(edge.edgeKey);
-      else edgeKeysByRatio.set(ratio, [edge.edgeKey]);
-      if (isTemplateLinkRatio(ratio)) templateEdgeCount += 1;
+      const decision = classifyTemplateLinkEdge(edge, { index, pagesFetched, placementAvailable, ubiquityAvailable });
+      observeTemplateLinkDetection(tally, edge, decision);
+      if (decision.source === SiteHealthLinkClassificationSources.placement) placementEdgeCount += 1;
+      if (decision.isTemplate) templateEdgeCount += 1;
+      if (!decision.isTemplate && decision.templateRatio == null) continue;
+      const key = writeGroupKey(decision.isTemplate, decision.templateRatio);
+      const group = edgeKeysByValue.get(key);
+      if (group) group.edgeKeys.push(decision.edgeKey);
+      else {
+        edgeKeysByValue.set(key, {
+          isTemplate: decision.isTemplate,
+          templateRatio: decision.templateRatio,
+          edgeKeys: [decision.edgeKey]
+        });
+      }
     }
   }
   const now = (/* @__PURE__ */ new Date()).toISOString();
-  for (const [ratio, edgeKeys] of edgeKeysByRatio) {
-    const isTemplate = isTemplateLinkRatio(ratio);
-    for (const chunk of chunked(edgeKeys)) {
-      db.update(siteCrawlEdges).set({ isTemplate, templateRatio: ratio, updatedAt: now }).where(and11(scopeFilter, inArray4(siteCrawlEdges.edgeKey, chunk))).run();
+  for (const group of edgeKeysByValue.values()) {
+    for (const chunk of chunked(group.edgeKeys)) {
+      db.update(siteCrawlEdges).set({ isTemplate: group.isTemplate, templateRatio: group.templateRatio, updatedAt: now }).where(and11(scopeFilter, inArray4(siteCrawlEdges.edgeKey, chunk))).run();
     }
   }
-  return { detection, templateEdgeCount };
+  return {
+    detection: templateLinkDetection({ placementAvailable, ubiquityAvailable, tally }),
+    templateEdgeCount,
+    placementEdgeCount
+  };
 }
 
 // src/execute-site-audit.ts
@@ -8230,6 +8268,11 @@ async function executeSiteAudit(db, runId, projectId, opts = {}) {
     for (const page of pages) persistPage(tx, page, now);
     bindReferences(tx, changedUrls, now);
   };
+  const edgePlacementColumns = (edge) => ({
+    placementNavigationOccurrences: edge.placementOccurrences?.navigation ?? null,
+    placementContentOccurrences: edge.placementOccurrences?.content ?? null,
+    placementUnknownOccurrences: edge.placementOccurrences?.unknown ?? null
+  });
   const persistEdge = (tx, edge, now) => {
     if (isSelfLink(edge.from, edge.to)) return;
     observedEdges.set(edge.key, edge);
@@ -8250,6 +8293,7 @@ async function executeSiteAudit(db, runId, projectId, opts = {}) {
       followableOccurrences: edge.followableOccurrences,
       nofollowOccurrences: edge.nofollowOccurrences,
       anchors: edge.anchorSummaries.map((anchor) => anchor.text),
+      ...edgePlacementColumns(edge),
       createdAt: now,
       updatedAt: now
     }).onConflictDoUpdate({
@@ -8266,6 +8310,7 @@ async function executeSiteAudit(db, runId, projectId, opts = {}) {
         followableOccurrences: edge.followableOccurrences,
         nofollowOccurrences: edge.nofollowOccurrences,
         anchors: edge.anchorSummaries.map((anchor) => anchor.text),
+        ...edgePlacementColumns(edge),
         updatedAt: now
       }
     }).run();
@@ -8373,16 +8418,20 @@ async function executeSiteAudit(db, runId, projectId, opts = {}) {
     if (crawlSummary.complete && !hasAuditedPages) {
       throw emptyCompleteCrawlError(crawlSummary.rootUrl, crawlSummary.finalRootUrl, crawlSummary.pagesObserved);
     }
+    const placementRulesetVersion = crawlSummary.linkPlacementRulesetVersion ?? null;
     const templateLinks = classifySiteCrawlTemplateLinks(
       db,
       { projectId, runId, attemptId },
-      crawlSummary.pagesFetched
+      crawlSummary.pagesFetched,
+      placementRulesetVersion
     );
     log12.info("template-links", {
       runId,
       projectId,
       detection: templateLinks.detection,
-      templateEdges: templateLinks.templateEdgeCount
+      templateEdges: templateLinks.templateEdgeCount,
+      placementEdges: templateLinks.placementEdgeCount,
+      placementRulesetVersion
     });
     const graphLayout = await prepareSiteCrawlGraphLayout(db, {
       projectId,
@@ -8497,6 +8546,7 @@ async function executeSiteAudit(db, runId, projectId, opts = {}) {
         deadLinksChecked,
         deadLinksFound,
         templateDetection: templateLinks.detection,
+        linkPlacementRulesetVersion: placementRulesetVersion,
         createdAt: finishedAt,
         updatedAt: finishedAt
       }).run();
@@ -9034,7 +9084,7 @@ function readStoredGroundingSources(rawResponse) {
   return result;
 }
 async function backfillInsightsCommand(project, opts) {
-  const { IntelligenceService: IntelligenceService2 } = await import("./intelligence-service-HKDIVTZB.js");
+  const { IntelligenceService: IntelligenceService2 } = await import("./intelligence-service-AFAFGLCM.js");
   const config = loadConfig();
   const db = createClient(config.database);
   migrate(db);