fix(fingerprinter): run built-in normalizers before user rules (#20)

User-defined normalizers ran first, so a broad rule like `/\d+/g` would
strip digits out of UUIDs and hex addresses before UUID_RE / HEX_RE could
match. Every trade ID and tx hash then produced a distinct fingerprint,
which turned each occurrence into a fresh onset and suppression never
kicked in.

Built-ins now collapse structural identifiers first, and user rules
compose on top of the normalized output. Adds regression tests covering
UUID and hex survival when a `\d+` user rule is configured.

Author: Royal-lobster

.changeset/normalizer-order.md

@@ -0,0 +1,15 @@
+---
+"@iqai/alert-logger": patch
+---
+
+fix(fingerprinter): run built-in normalizers before user-defined ones
+
+User-defined normalizers previously ran before the built-in ones, so a
+broad rule like `{ pattern: /\d+/g, replacement: "<num>" }` would strip
+digits out of UUIDs and hex addresses before `UUID_RE` and `HEX_RE` had a
+chance to match. Every trade ID or transaction hash then produced a
+distinct fingerprint, which made the aggregator treat each occurrence as
+a fresh onset and suppression never kicked in.
+
+Built-ins now collapse structural identifiers first, and user rules
+compose on top of the normalized output.

src/core/fingerprinter.test.ts

@@ -118,7 +118,7 @@ describe('fingerprint', () => {
   // ── Custom normalizers ──────────────────────────────────────────────
 
   describe('custom normalizers', () => {
-    it('applies user-defined normalizers before builtins', () => {
+    it('applies user-defined normalizers on top of builtins', () => {
       const custom: FingerprintConfig = {
         stackDepth: 3,
         normalizers: [{ pattern: /order-\w+/g, replacement: '<order>' }],
@@ -128,16 +128,48 @@ describe('fingerprint', () => {
       expect(a).toBe(b)
     })
 
-    it('user normalizers run before builtins so they can match raw text', () => {
+    it('does not let broad user rules break built-in UUID normalization', () => {
+      // A user-supplied `/\d+/g` rule used to run before built-ins and strip
+      // digits out of UUIDs, leaving distinct strings where a UUID should
+      // have collapsed to "<uuid>". Built-ins now run first, so structural
+      // identifiers survive.
       const custom: FingerprintConfig = {
         stackDepth: 3,
-        normalizers: [{ pattern: /ID:\d+/g, replacement: '<id>' }],
+        normalizers: [{ pattern: /\d+/g, replacement: '<num>' }],
       }
-      // The user normalizer matches "ID:42" before the builtin number normalizer
-      // could turn "42" into "<num>"
-      const hash = fingerprint('E', 'lookup ID:42', undefined, custom)
-      const hashWithDifferentId = fingerprint('E', 'lookup ID:99', undefined, custom)
-      expect(hash).toBe(hashWithDifferentId)
+      const a = fingerprint(
+        'E',
+        'Trade a51c80e4-3307-4d5f-a035-03c8fd1f767d permanently failed',
+        undefined,
+        custom,
+      )
+      const b = fingerprint(
+        'E',
+        'Trade 1bbc368f-a1fa-44af-9d04-ffaa804a30b1 permanently failed',
+        undefined,
+        custom,
+      )
+      expect(a).toBe(b)
+    })
+
+    it('does not let broad user rules break built-in hex normalization', () => {
+      const custom: FingerprintConfig = {
+        stackDepth: 3,
+        normalizers: [{ pattern: /\d+/g, replacement: '<num>' }],
+      }
+      const a = fingerprint(
+        'E',
+        'safeTxHash=0x9db6e277fdea4e17ef1597e358d7bd893d257cf62378180a9d279fdeb1d0ab58',
+        undefined,
+        custom,
+      )
+      const b = fingerprint(
+        'E',
+        'safeTxHash=0xf51b9f89c5919f9efffd9eb2450251bf668ec8410aa5c599c267027fc92b8466',
+        undefined,
+        custom,
+      )
+      expect(a).toBe(b)
     })
   })
 

src/core/fingerprinter.ts

@@ -16,13 +16,16 @@ const BUILTIN_NORMALIZERS: NormalizerRule[] = [
 function normalizeMessage(message: string, userNormalizers: NormalizerRule[]): string {
   let result = message
 
-  // Apply user-defined normalizers first
-  for (const rule of userNormalizers) {
+  // Apply built-in normalizers first so structural identifiers (UUIDs, hex
+  // addresses, timestamps, numbers) are collapsed before user rules run.
+  // Running user rules first lets a broad pattern like `/\d+/g` strip digits
+  // out of UUIDs/hex, which prevents the structural regexes from matching
+  // and turns every ID into its own fingerprint.
+  for (const rule of BUILTIN_NORMALIZERS) {
     result = result.replace(rule.pattern, rule.replacement)
   }
 
-  // Then apply built-in normalizers
-  for (const rule of BUILTIN_NORMALIZERS) {
+  for (const rule of userNormalizers) {
     result = result.replace(rule.pattern, rule.replacement)
   }