add origin

src/plugin.ts

@@ -477,9 +477,21 @@ export function videoSubtitleKeys(produced: ProducedSubtitle[]): Record<string,
     return keys;
 }
 
-/** Keys on the subtitle file's own record. */
+/**
+ * Keys on the subtitle file's own record.
+ *
+ * `source/extract` (METADATA_KEYS.md §5) says the bytes came out of a video
+ * container. The video record already implies it through `extractedSubtitles/`,
+ * but that is the wrong side of the relation to store it on: a subtitle fetched
+ * over the swarm by CID arrives WITHOUT its video record, so a reader holding
+ * only the file would have no way to tell an extracted track from a sidecar
+ * someone dropped next to the video.
+ */
 export function subtitleRecordKeys(videoCid: string, lang3: string): Record<string, string> {
-    const keys: Record<string, string> = { [`videos/${videoCid}`]: 'true' };
+    const keys: Record<string, string> = {
+        [`videos/${videoCid}`]: 'true',
+        'source/extract': 'true',
+    };
     if (lang3 !== 'und') keys.subtitleLanguage = lang3;
     return keys;
 }
@@ -499,7 +511,14 @@ export function legacySubtitleMigration(existingMeta: Record<string, unknown> |
     const langs = legacyCsvMembers(existingMeta, 'subtitleLanguages');
     if (langs) {
         deletes.push('subtitleLanguages');
-        for (const l of langs.map(toLang3)) if (l !== 'und') sets[`subtitleLanguages/${l}`] = 'true';
+        for (const l of langs.map(toLang3)) {
+            if (l === 'und') continue;
+            sets[`subtitleLanguages/${l}`] = 'true';
+            // …and the union (§9 rule #7). The live write path below already
+            // does this; the migration path did not, so records that only ever
+            // went through it were left half-written.
+            sets[`languages/${l}`] = 'true';
+        }
     }
 
     const subs = legacyCsvMembers(existingMeta, 'subtitles');

test/extractor.test.ts

@@ -284,9 +284,16 @@ describe('record keys (METADATA_KEYS.md §8)', () => {
         expect(keys.every((k) => k.split('/').length === 3)).toBe(true);
     });
 
-    it('subtitle record carries the back-pointer and a determined language only', () => {
+    it('subtitle record carries the back-pointer, its origin, and a determined language only', () => {
-        expect(subtitleRecordKeys('bagvideo', 'fre')).toEqual({ 'videos/bagvideo': 'true', subtitleLanguage: 'fre' });
+        expect(subtitleRecordKeys('bagvideo', 'fre')).toEqual({
-        expect(subtitleRecordKeys('bagvideo', 'und')).toEqual({ 'videos/bagvideo': 'true' });
+            'videos/bagvideo': 'true',
+            'source/extract': 'true',
+            subtitleLanguage: 'fre',
+        });
+        expect(subtitleRecordKeys('bagvideo', 'und')).toEqual({
+            'videos/bagvideo': 'true',
+            'source/extract': 'true',
+        });
     });
 });
 
@@ -310,7 +317,13 @@ describe('legacySubtitleMigration', () => {
             subtitles: 'cidS',
         })).toEqual({
             deletes: ['extractedSubtitles', 'subtitleLanguages', 'subtitles'],
-            sets: { 'subtitleLanguages/eng': 'true', 'subtitleLanguages/fre': 'true', 'subtitles/und/cidS': 'true' },
+            sets: {
+                'subtitleLanguages/eng': 'true',
+                'languages/eng': 'true',
+                'subtitleLanguages/fre': 'true',
+                'languages/fre': 'true',
+                'subtitles/und/cidS': 'true',
+            },
         });
     });
 
@@ -495,8 +508,17 @@ describe.skipIf(!ffmpegReady)('process — end to end against a fake core', () =
         expect(video.subtitleLanguages).toBeUndefined();
         expect(video['subtitleLanguages/und']).toBeUndefined();
 
-        expect(core.records.get(fre)).toEqual({ [`videos/${VIDEO_CID}`]: 'true', subtitleLanguage: 'fre' });
+        // `source/extract` rides on the subtitle's OWN record, so a peer that
-        expect(core.records.get(und)).toEqual({ [`videos/${VIDEO_CID}`]: 'true' });
+        // fetched only the file can still tell it came out of a container.
+        expect(core.records.get(fre)).toEqual({
+            [`videos/${VIDEO_CID}`]: 'true',
+            'source/extract': 'true',
+            subtitleLanguage: 'fre',
+        });
+        expect(core.records.get(und)).toEqual({
+            [`videos/${VIDEO_CID}`]: 'true',
+            'source/extract': 'true',
+        });
 
         // The video PATCH carried only new keys — nothing the record already had.
         const patch = writesTo(core, VIDEO_CID).find((c) => c.method === 'PATCH')!;