fix(matching): guard tag stripping against real-title false positives

Address code-review findings on the tag-stripping rules:

- underscore suffix is now vocabulary-gated, so "Mr_Robot",
  "Cowboy_Bebop", "Mrs_Davis" keep their second word
- leading single-segment tags before a spaced dash stay 2-3 chars
  (only hyphen-compounds like "4K-DE" and pipe-tags like "EXYU|" may be
  wider), so "DUNE - Part Two" and "ALIEN - Covenant" are left intact
- "IN" is excluded from the weak joined-dash/underscore paths so
  "drive-in" and "Plug-in" are not truncated (India still strips via
  the strong "IN| " / "IN - " forms)

The display-side stripCountryPrefix() mirrors the narrowed dash rule.
Corpus coverage is unchanged at 99% (245/248); new counter-example
tests lock in the guards.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

libs/shared/interfaces/src/lib/title-normalization.util.spec.ts

@@ -83,6 +83,11 @@ describe('provider tag stripping', () => {
         );
     });
 
+    it('keeps bare 4-5 char words before a spaced dash (real titles)', () => {
+        expect(normalizeTitle('DUNE - Part Two')).toBe('dune part two');
+        expect(normalizeTitle('ALIEN - Covenant')).toBe('alien covenant');
+    });
+
     it('strips underscore and double-dash suffix tags', () => {
         expect(normalizeTitle('Fallout_eng')).toBe('fallout');
         expect(normalizeTitle('Breaking Bad (US)_msub')).toBe('breaking bad');
@@ -94,6 +99,12 @@ describe('provider tag stripping', () => {
         expect(normalizeTitle('The_Last_of_Us')).toBe('the last of us');
     });
 
+    it('keeps sole-underscore titles whose tail is not a known tag', () => {
+        expect(normalizeTitle('Mr_Robot')).toBe('mr robot');
+        expect(normalizeTitle('Cowboy_Bebop')).toBe('cowboy bebop');
+        expect(normalizeTitle('Mrs_Davis')).toBe('mrs davis');
+    });
+
     it('strips joined dash tags only for case-uniform vocabulary tokens', () => {
         expect(normalizeTitle('Breaking Bad-eng')).toBe('breaking bad');
         expect(normalizeTitle('The Last of Us-DE')).toBe('the last of us');
@@ -102,6 +113,11 @@ describe('provider tag stripping', () => {
         expect(normalizeTitle('Kick-It')).toBe('kick it');
     });
 
+    it('keeps English hyphenated word endings that collide with codes', () => {
+        expect(normalizeTitle('drive-in')).toBe('drive in');
+        expect(normalizeTitle('Plug-in')).toBe('plug in');
+    });
+
     it('strips bare trailing UPPERCASE vocabulary tags', () => {
         expect(normalizeTitle('The Pitt (2025) DE')).toBe('the pitt');
         expect(normalizeTitle('Breaking Bad ES')).toBe('breaking bad');

libs/shared/interfaces/src/lib/title-normalization.util.ts

@@ -37,15 +37,27 @@ const WRAPPED_TAG_PREFIX = /^\s*\|(?=[0-9+]*[A-Z])[A-Z0-9+]{2,5}\|\s*/;
 /**
  * Leading channel/language prefix like "EN - ", "DE| ", "FR: ", including
  * compound provider/quality forms ("4K-DE - ", "AR-SUBS - ", "4K-OSN+ - ")
- * and longer tags ("EXYU| ", "MULTI| ").
+ * and longer pipe-tagged forms ("EXYU| ", "MULTI| ").
  * UPPERCASE-only on purpose: a case-insensitive match would amputate real
  * title words ("It: Chapter Two" → "Chapter Two"). Every segment must
  * contain a letter so numeric titles ("1917 - ...") are never tags.
- * Colon separators stay limited to 2–3 chars — longer acronyms before a
- * colon are franchise titles ("NCIS: LA"), not tags.
+ *
+ * Separator strength gates how wide a single segment may be:
+ *   - dash ("EN - "): compound OR 2–3 chars — a bare 4–5 char word before
+ *     a spaced dash is a real title ("DUNE - Part Two", "ALIEN - Covenant")
+ *   - pipe ("EXYU| "): compound OR 2–5 chars — a pipe is a strong tag signal
+ *   - colon ("EN: "): 2–3 chars — longer acronyms are franchise titles
+ *     ("NCIS: LA")
  */
-const LANGUAGE_PREFIX =
-    /^(?:(?=[0-9+]*[A-Z])[A-Z0-9+]{2,5}(?:-(?=[0-9+]*[A-Z])[A-Z0-9+]{2,6}){0,2}\s*[-|]\s+|(?=[0-9+]*[A-Z])[A-Z0-9+]{2,3}\s*:\s+)/;
+const SEG = '(?=[0-9+]*[A-Z])[A-Z0-9+]';
+const COMPOUND_TAG = `${SEG}{2,5}(?:-${SEG}{2,6}){1,2}`;
+const LANGUAGE_PREFIX = new RegExp(
+    '^(?:' +
+        `(?:${COMPOUND_TAG}|${SEG}{2,3})\\s*-\\s+` +
+        `|(?:${COMPOUND_TAG}|${SEG}{2,5})\\s*\\|\\s+` +
+        `|${SEG}{2,3}\\s*:\\s+` +
+        ')'
+);
 
 /**
  * Curated whitelist for TRAILING language/subtitle tags ("Fallout_eng",
@@ -61,8 +73,16 @@ const TRAILING_TAG_VOCABULARY = new Set([
     'SC', 'SE', 'SUB', 'SUBS', 'SW', 'TA', 'TL', 'TR', 'TUR',
 ]);
 
+/**
+ * Vocabulary tags that are also common English hyphenated-word endings
+ * ("drive-in", "plug-in"). Accepted only after a STRONG separator (bare
+ * spaced/uppercase form), never in the weak joined-dash/underscore forms
+ * where "X-in"/"X_in" reads as a title, not a tag.
+ */
+const WEAK_JOIN_EXCLUSIONS = new Set(['IN']);
+
 const DOUBLE_DASH_SUFFIX = /[-–]{2}[A-Za-z]{2,5}\s*$/;
-const UNDERSCORE_SUFFIX = /_[A-Za-z]{2,5}\s*$/;
+const UNDERSCORE_SUFFIX = /_([A-Za-z]{2,5})\s*$/;
 const JOINED_DASH_SUFFIX = /-([A-Za-z]{2,5})\s*$/;
 const TRAILING_TAG_SUFFIX = /\s([A-Z]{2,5})\s*$/;
 
@@ -71,6 +91,21 @@ function isCaseUniform(token: string): boolean {
     return token === token.toLowerCase() || token === token.toUpperCase();
 }
 
+/**
+ * A captured joined-dash/underscore token is provider metadata only when
+ * it is a known vocabulary tag, case-uniform, and not one of the English
+ * word-forming exclusions. This keeps "Mr_Robot", "Cowboy_Bebop",
+ * "drive-in", and "Plug-in" intact while stripping "_eng", "-DE", "-it".
+ */
+function isJoinedTag(token: string): boolean {
+    const upper = token.toUpperCase();
+    return (
+        TRAILING_TAG_VOCABULARY.has(upper) &&
+        !WEAK_JOIN_EXCLUSIONS.has(upper) &&
+        isCaseUniform(token)
+    );
+}
+
 /** ES2015-safe trailing-whitespace trim (the lib target predates trimEnd). */
 function trimRight(value: string): string {
     return value.replace(/\s+$/, '');
@@ -103,22 +138,22 @@ function stripTrailingTagOnce(value: string): string {
         return trimRight(result.replace(DOUBLE_DASH_SUFFIX, ''));
     }
 
-    // "Fallout_eng" — but not "The_Last_of_Us", where underscores are
-    // space substitutes (only strip when this is the sole underscore).
+    // "Fallout_eng" — but not "The_Last_of_Us" (underscores as spaces, only
+    // strip a sole underscore) and not "Mr_Robot"/"Cowboy_Bebop" (the tail
+    // must be a known tag, so the segment is real-title evidence otherwise).
+    const underscore = result.match(UNDERSCORE_SUFFIX);
     if (
-        UNDERSCORE_SUFFIX.test(result) &&
-        result.indexOf('_') === result.lastIndexOf('_')
+        underscore &&
+        result.indexOf('_') === result.lastIndexOf('_') &&
+        isJoinedTag(underscore[1])
     ) {
         return trimRight(result.replace(UNDERSCORE_SUFFIX, ''));
     }
 
+    // "Breaking Bad-eng", "The Last of Us-DE" — but not "drive-in"/"Plug-in".
+    // hasLowercase gates ALL-CAPS titles out (no casing signal to trust).
     const joined = result.match(JOINED_DASH_SUFFIX);
-    if (
-        joined &&
-        TRAILING_TAG_VOCABULARY.has(joined[1].toUpperCase()) &&
-        isCaseUniform(joined[1]) &&
-        (joined[1] !== joined[1].toUpperCase() || hasLowercase)
-    ) {
+    if (joined && hasLowercase && isJoinedTag(joined[1])) {
         return trimRight(result.replace(JOINED_DASH_SUFFIX, ''));
     }
 

libs/shared/m3u-utils/src/lib/strip-country-prefix.util.spec.ts

@@ -53,9 +53,9 @@ describe('stripCountryPrefix', () => {
             );
         });
 
-        it('strips longer uppercase tags', () => {
-            expect(stripCountryPrefix('EXYU - News')).toBe('News');
-            expect(stripCountryPrefix('MULTI - Movies')).toBe('Movies');
+        it('strips longer pipe-tagged prefixes', () => {
+            expect(stripCountryPrefix('EXYU| News')).toBe('News');
+            expect(stripCountryPrefix('MULTI| Movies')).toBe('Movies');
         });
 
         it('never treats numeric fragments as tags', () => {
@@ -64,6 +64,15 @@ describe('stripCountryPrefix', () => {
             );
         });
 
+        it('keeps bare 4-5 char words before a spaced dash (real titles)', () => {
+            expect(stripCountryPrefix('DUNE - Part Two')).toBe(
+                'DUNE - Part Two'
+            );
+            expect(stripCountryPrefix('ALIEN - Covenant')).toBe(
+                'ALIEN - Covenant'
+            );
+        });
+
         it('only strips the first tag segment', () => {
             expect(stripCountryPrefix('ES - A3 - Sports')).toBe('A3 - Sports');
         });

libs/shared/m3u-utils/src/lib/strip-country-prefix.util.ts

@@ -4,16 +4,17 @@
  * ("Sky - Sports F1", "Discovery - Science"). Pipes are conventionally
  * used only as tag separators, so they always count.
  *
- * A tag is 1–3 short UPPERCASE alphanumeric segments ("US", "4K-DE",
- * "AR-SUBS", "OSN+"). Every segment must contain a letter so numeric
- * titles ("1917 - ...") are never treated as tags. Colon separators stay
- * limited to plain 2–3 char tags — longer acronyms before a colon are
- * franchise titles ("NCIS: LA"), not tags.
+ * Before a dash a tag is either a compound ("4K-DE", "AR-SUBS", "4K-OSN+"
+ * — the inner hyphen is the tag signal) or a plain 2–3 char code ("US",
+ * "4K"). A bare 4–5 char word before a spaced dash is a real title
+ * ("DUNE - Part Two", "ALIEN - Covenant"), so it is not a tag. Colon tags
+ * stay 2–3 chars — longer acronyms are franchise titles ("NCIS: LA").
+ * Every segment must contain a letter so numbers ("1917 - ...") are safe.
  */
 const DASH_SEPARATORS = [' - ', '- ', ' -'];
 const COLON_SEPARATOR = ': ';
 const TAG_PREFIX_PATTERN =
-    /^(?=[0-9+]*[A-Z])[A-Z0-9+]{2,5}(?:-(?=[0-9+]*[A-Z])[A-Z0-9+]{2,6}){0,2}$/;
+    /^(?:(?=[0-9+]*[A-Z])[A-Z0-9+]{2,5}(?:-(?=[0-9+]*[A-Z])[A-Z0-9+]{2,6}){1,2}|(?=[0-9+]*[A-Z])[A-Z0-9+]{2,3})$/;
 const COLON_TAG_PATTERN = /^(?=[0-9+]*[A-Z])[A-Z0-9+]{2,3}$/;
 
 interface SeparatorMatch {