apps/worker/src/lib/compress.ts

@@ -474,12 +474,23 @@ export function compressSnapshots(
     ...(hypotheses.length ? hypotheses.map((h) => `- ${h}`) : ["- (none)"]),
   ].join("\n");
 
+  // Fingerprint inputs are intentionally coarse so that recurring problems
+  // cluster into a single Insight row instead of producing a new row per
+  // user/vehicle/category. Two prior issues this guards against:
+  //   1. URL carries vehicleId/categoryId UUIDs — every user produces a unique
+  //      URL, so identical kategori-bouncing sessions never deduplicate.
+  //      normalizePath collapses UUIDs/ULIDs/CUIDs/numeric ids to ":id".
+  //   2. header.severity is derived from rage-click count and varies between
+  //      P1/P2/P3 for the same root cause — it should be a property of the
+  //      Insight, not part of its identity. Dropped from the hash.
+  // Tag set + normalized path + first-error + first-failed-endpoint give
+  // enough discrimination because distinct UX failures already carry distinct
+  // tagger tags (vin_decode_*, payment_*, search_validation_*, etc.).
   const fingerprint = fingerprintHash([
     [...header.tags].sort().join(","),
-    url,
+    normalizePath(url),
     errors[0] ? normalizeError(errors[0]) : null,
-    failedEndpoints[0] ?? null,
-    header.severity,
+    failedEndpoints[0] ? normalizePath(failedEndpoints[0]) : null,
   ]);
 
   return {
@@ -511,6 +522,47 @@ function stripQuery(u: string): string {
   }
 }
 
+/**
+ * Collapse ID-like path segments to `:id` so URLs with embedded
+ * UUID/ULID/CUID/numeric identifiers fingerprint identically across users.
+ *
+ * Conservative on purpose — only matches segments whose entire content is a
+ * recognised id format, leaving meaningful path words alone. Used both for
+ * the page url and for failed-endpoint paths.
+ *
+ * Examples:
+ *   /dashboard/vehicles/6a3e7c44-d64b-4245-abc7-adb692a90fff/categories/29ee50e7-887f-4124-b1fc-bbeea5e358d1
+ *     → /dashboard/vehicles/:id/categories/:id
+ *   /api/orders/12345/items   → /api/orders/:id/items
+ *   /insights/i/cmpwrb82s002f14fza8lbc7f4 → /insights/i/:id
+ */
+function normalizePath(u: string | null | undefined): string | null {
+  if (!u) return u ?? null;
+  let path: string;
+  try {
+    const parsed = new URL(u, "https://x");
+    path = parsed.pathname || u;
+  } catch {
+    const q = u.indexOf("?");
+    path = q === -1 ? u : u.slice(0, q);
+  }
+  return path
+    .split("/")
+    .map((seg) => {
+      if (!seg) return seg;
+      // UUID v1–v8 canonical form
+      if (/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(seg)) return ":id";
+      // ULID — Crockford base32, exactly 26 chars
+      if (/^[0-9A-HJKMNP-TV-Z]{26}$/.test(seg)) return ":id";
+      // CUID2 / similar — lowercase alphanumeric, length >= 20 starting with a letter
+      if (/^[a-z][a-z0-9]{19,}$/.test(seg)) return ":id";
+      // Pure numeric id
+      if (/^\d+$/.test(seg)) return ":id";
+      return seg;
+    })
+    .join("/");
+}
+
 function truncate(s: string, n: number): string {
   return s.length > n ? `${s.slice(0, n)}…` : s;
 }