fix(insights): propagate rrweb counters & resolve click targets

Three coupled fixes that together stop frustrated sessions from being
mislabelled as `power_user_path` and make click timelines actually
diagnosable.

1) Counter propagation (compress.ts, compress-sessions.ts):
   compress now returns rageClickCount / network5xxCount / network4xxCount
   and writes them back to SessionMeta. Previously these stayed at 0
   forever because PostHog recording metadata doesn't expose them and
   nothing updated the row after compress ran.

2) Tag rule hardening (tagger.ts, compress-sessions.ts):
   - power_user_path v1 fallback now requires rageClickCount===0; without
     this guard any auth user with 20+ clicks (rage clusters included)
     was labelled a power user.
   - New frustrated_session tag (rageClickCount>=3, P2) for sustained
     friction beyond a single cluster.
   - compress-sessions re-runs tagSession+scoreSession after writing the
     fresh counters, so the corrected tags land on the row.

3) Click target enrichment (compress.ts):
   describeTarget used to read tagName/attributes off MouseInteraction
   events — fields that don't exist on rrweb type=3 source=2 — so every
   click rendered as `[el]`. Now compress builds a node map from
   FullSnapshot (type=2) and keeps it in sync with mutations
   (type=3 source=0), then resolves clicks via `d.id`. Clicks now show
   real tag + visible text / aria-label / name / data-testid.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/worker/src/jobs/compress-sessions.ts

@@ -3,7 +3,8 @@ import { getSnapshotSources, getSnapshotBlob, isConfigured } from "../lib/postho
 import { compressSnapshots, parseSnapshotBlob } from "../lib/compress";
 import { putText } from "../lib/minio";
 import { loadEnrichment } from "../lib/enrich";
-import { getCachedGroup } from "../lib/posthog-cache";
+import { getCachedGroup, getCachedPerson } from "../lib/posthog-cache";
+import { tagSession, scoreSession } from "../lib/tagger";
 import { alertSanitizationAnomaly } from "../lib/telegram";
 
 const PANEL_URL = process.env.PANEL_PUBLIC_URL ?? "https://sp.semih.ai";
@@ -41,18 +42,18 @@ export async function runCompressSessions(): Promise<{ compressed: number; faile
         events.push(...parseSnapshotBlob(blob));
       }
       const { customEvents } = await loadEnrichment(s.id);
+      const group = s.groupKey
+        ? await getCachedGroup("company", s.groupKey).catch(() => null)
+        : null;
       let companyContext: string | null = null;
-      if (s.groupKey) {
+      if (group) {
-        const group = await getCachedGroup("company", s.groupKey).catch(() => null);
+        const p = group.properties;
-        if (group) {
+        const parts: string[] = [];
-          const p = group.properties;
+        if (p.tier) parts.push(`tier=${p.tier}`);
-          const parts: string[] = [];
+        if (typeof p.mrr_usd === "number") parts.push(`mrr=$${p.mrr_usd}`);
-          if (p.tier) parts.push(`tier=${p.tier}`);
+        if (typeof p.seat_count === "number") parts.push(`seats=${p.seats_used ?? "?"}/${p.seat_count}`);
-          if (typeof p.mrr_usd === "number") parts.push(`mrr=$${p.mrr_usd}`);
+        if (p.industry) parts.push(`industry=${p.industry}`);
-          if (typeof p.seat_count === "number") parts.push(`seats=${p.seats_used ?? "?"}/${p.seat_count}`);
+        companyContext = parts.join(", ") || null;
-          if (p.industry) parts.push(`industry=${p.industry}`);
-          companyContext = parts.join(", ") || null;
-        }
       }
 
       const out = compressSnapshots(
@@ -113,9 +114,42 @@ export async function runCompressSessions(): Promise<{ compressed: number; faile
         },
       });
 
+      // Re-tag with the freshly-derived rrweb counters. The tag step ran
+      // before compress and saw rageClickCount=0 etc., so rules like
+      // ux_friction/frustrated_session and the rage-guarded power_user_path
+      // fallback need a second pass once the real counts are known.
+      const updatedMeta = {
+        ...s,
+        rageClickCount: out.rageClickCount,
+        network5xxCount: out.network5xxCount,
+        network4xxCount: out.network4xxCount,
+      };
+      const personCache = s.posthogDistinctId
+        ? await getCachedPerson(s.posthogDistinctId).catch(() => null)
+        : null;
+      const retagCtx = {
+        customEvents,
+        userProperties: personCache?.properties ?? {},
+        groupProperties: group?.properties ?? null,
+      };
+      const retag = tagSession(updatedMeta, retagCtx);
+      const rescore = scoreSession(updatedMeta, retag.severity, 0.5, 0.2, retagCtx.groupProperties);
+
       await prisma.sessionMeta.update({
         where: { id: s.id },
-        data: { status: "compressed", fingerprint: out.fingerprint, processedAt: new Date() },
+        data: {
+          status: "compressed",
+          fingerprint: out.fingerprint,
+          processedAt: new Date(),
+          // Backfill rrweb-derived counters that ingest can't see (PostHog
+          // recording metadata doesn't expose rage / 4xx / 5xx counts).
+          rageClickCount: out.rageClickCount,
+          network5xxCount: out.network5xxCount,
+          network4xxCount: out.network4xxCount,
+          tags: retag.tags,
+          severity: retag.severity,
+          score: rescore,
+        },
       });
 
       if (anomaly) {

apps/worker/src/lib/compress.ts

@@ -12,6 +12,11 @@ export type CompressedOutput = {
   sanitization: SanitizationReport;
   keyEvents: string[];
   hypotheses: string[];
+  // Rrweb-derived counters propagated back to SessionMeta so the tagger can
+  // see signals (e.g. rage clusters) that aren't available at ingest time.
+  rageClickCount: number;
+  network5xxCount: number;
+  network4xxCount: number;
 };
 
 type RREvent = {
@@ -77,6 +82,65 @@ export function compressSnapshots(
   const errors: string[] = [];
   const network5xx: string[] = [];
   const failedEndpoints: string[] = [];
+  let rageClickCount = 0;
+  let network4xxCount = 0;
+
+  // rrweb node id → element info, populated from FullSnapshot (type=2) and kept
+  // in sync with mutations (type=3 source=0). MouseInteraction/Input events
+  // (type=3 source=2|5) only carry the rrweb `id` — without this map every
+  // click resolves to bare "[el]" and all UI-friction analysis loses signal.
+  type NodeInfo = {
+    nodeType?: number;
+    tagName?: string;
+    attrs?: Record<string, string>;
+    textContent?: string;
+    childIds: number[];
+  };
+  const nodeMap = new Map<number, NodeInfo>();
+
+  const indexNode = (n: any): void => {
+    if (!n || typeof n.id !== "number") return;
+    const info: NodeInfo = { nodeType: n.type, childIds: [] };
+    if (n.type === 2) {
+      // ELEMENT
+      info.tagName = String(n.tagName ?? "").toLowerCase();
+      if (n.attributes && typeof n.attributes === "object") info.attrs = n.attributes;
+    } else if (n.type === 3) {
+      // TEXT
+      info.textContent = typeof n.textContent === "string" ? n.textContent : "";
+    }
+    nodeMap.set(n.id, info);
+    if (Array.isArray(n.childNodes)) {
+      for (const c of n.childNodes) {
+        indexNode(c);
+        if (c && typeof c.id === "number") info.childIds.push(c.id);
+      }
+    }
+  };
+
+  const resolveText = (id: number, depth = 0): string => {
+    if (depth > 3) return "";
+    const node = nodeMap.get(id);
+    if (!node) return "";
+    if (node.nodeType === 3) return node.textContent ?? "";
+    let out = "";
+    for (const c of node.childIds) {
+      out += resolveText(c, depth + 1);
+      if (out.length > 60) break;
+    }
+    return out;
+  };
+
+  const describeTarget = (d: any): string => {
+    const id = typeof d?.id === "number" ? d.id : null;
+    if (id === null) return "[el]";
+    const node = nodeMap.get(id);
+    if (!node || !node.tagName) return "[el]";
+    const attrs = node.attrs ?? {};
+    const label = (resolveText(id) || attrs["aria-label"] || attrs["name"] || attrs["data-testid"] || "").trim();
+    const text = truncate(label, 20);
+    return `[${node.tagName}${text ? `:"${text}"` : ""}]`;
+  };
   let url: string | null = header.startUrl;
 
   const formatCustom = (ev: CanonicalEvent): string => {
@@ -139,6 +203,11 @@ export function compressSnapshots(
     const tRel = Math.max(0, Math.round((ev.timestamp - t0) / 1000));
     const ts = formatTs(tRel);
 
+    if (ev.type === 2) {
+      // FullSnapshot — seed the node map with the current DOM tree.
+      indexNode(ev.data?.node);
+      continue;
+    }
     if (ev.type === 4) {
       // Meta — URL changes
       const href = ev.data?.href;
@@ -164,6 +233,8 @@ export function compressSnapshots(
           if (status >= 500) {
             network5xx.push(`${method} ${rurl}`);
             failedEndpoints.push(rurl);
+          } else if (status >= 400) {
+            network4xxCount++;
           }
         }
       }
@@ -172,7 +243,42 @@ export function compressSnapshots(
     if (ev.type === 3) {
       const d = ev.data ?? {};
       const source = d.source;
-      // 2 = MouseInteraction, 5 = Input, 6 = Scroll, 9 = ViewportResize
+      // 0 = Mutation, 2 = MouseInteraction, 5 = Input, 6 = Scroll, 9 = ViewportResize
+      if (source === 0) {
+        // Keep node map in sync. We don't process removes (a stale entry only
+        // matters if a later click targets it, which can't happen for removed
+        // nodes); same for moves (parent re-link doesn't change label).
+        if (Array.isArray(d.adds)) {
+          for (const a of d.adds) {
+            indexNode(a?.node);
+            const parent = typeof a?.parentId === "number" ? nodeMap.get(a.parentId) : null;
+            if (parent && a?.node && typeof a.node.id === "number") {
+              parent.childIds.push(a.node.id);
+            }
+          }
+        }
+        if (Array.isArray(d.attributes)) {
+          for (const a of d.attributes) {
+            if (typeof a?.id !== "number") continue;
+            const node = nodeMap.get(a.id);
+            if (!node) continue;
+            node.attrs = node.attrs ?? {};
+            for (const [k, v] of Object.entries(a.attributes ?? {})) {
+              if (v === null) delete node.attrs[k];
+              else node.attrs[k] = String(v);
+            }
+          }
+        }
+        if (Array.isArray(d.texts)) {
+          for (const t of d.texts) {
+            if (typeof t?.id !== "number") continue;
+            const node = nodeMap.get(t.id);
+            if (!node) continue;
+            node.textContent = typeof t.value === "string" ? t.value : "";
+          }
+        }
+        continue;
+      }
       if (source === 2) {
         const target = describeTarget(d);
         const now = ev.timestamp;
@@ -181,6 +287,7 @@ export function compressSnapshots(
           if (consecutiveClicks >= 3 && !rageEmitted) {
             lines.push(`${ts} → 😡 rage_click ${target} (${consecutiveClicks} in <1.5s)`);
             rageEmitted = true;
+            rageClickCount++;
           }
         } else {
           if (consecutiveClicks > 0 && !rageEmitted) {
@@ -348,6 +455,9 @@ export function compressSnapshots(
     sanitization: report,
     keyEvents,
     hypotheses,
+    rageClickCount,
+    network5xxCount: network5xx.length,
+    network4xxCount,
   };
 }
 
@@ -367,13 +477,6 @@ function stripQuery(u: string): string {
   }
 }
 
-function describeTarget(d: any): string {
-  const tag = d?.tagName?.toLowerCase?.() ?? "el";
-  const attrs = d?.attributes ?? {};
-  const text = truncate(String(d?.text ?? attrs?.["aria-label"] ?? attrs?.name ?? ""), 20);
-  return `[${tag}${text ? `:"${text}"` : ""}]`;
-}
-
 function truncate(s: string, n: number): string {
   return s.length > n ? `${s.slice(0, n)}…` : s;
 }

apps/worker/src/lib/tagger.ts

@@ -62,6 +62,14 @@ export function tagSession(s: SessionMeta, ctx: EnrichmentCtx | null): TagResult
     severity = bump(severity, "P2");
   }
 
+  // ─── Frustrated session ───
+  // Sustained rage clicking (3+ clusters) is a stronger signal than a single
+  // cluster — promote it past ux_friction so it surfaces above generic noise.
+  if (s.rageClickCount >= 3) {
+    tags.push("frustrated_session");
+    severity = bump(severity, "P2");
+  }
+
   // ─── VIN decode failure pattern ───
   const vinFails = events.filter((e) => e.name === "vin_decode_failed");
   if (vinFails.length >= 2) {
@@ -176,12 +184,15 @@ export function tagSession(s: SessionMeta, ctx: EnrichmentCtx | null): TagResult
       tags.push("power_user_path");
     }
   }
-  // v1 fallback (rrweb-only): authenticated + many clicks + no errors and no other tag yet
+  // v1 fallback (rrweb-only): authenticated + many clicks + no errors/rage and no other tag yet.
+  // Without the rage-click guard this fired on frustrated users (20+ clicks
+  // including rage clusters → mislabelled as power users).
   if (
     !tags.length &&
     s.isAuthenticated &&
     s.clickCount >= 20 &&
-    s.errorCount === 0
+    s.errorCount === 0 &&
+    s.rageClickCount === 0
   ) {
     tags.push("power_user_path");
   }