experiments in a post-browser web
fix(entities): remove all regex-extracted places in migration (too noisy)
+5
-3
+5
-3
backend/electron/datastore.ts
+5
-3
backend/electron/datastore.ts
···
1731
1731
function migrateLowQualityEntities(): void {
1732
1732
if (!db) return;
1733
1733
1734
-
const MIGRATION_ID = 'remove_low_quality_entities_v1';
1734
+
const MIGRATION_ID = 'remove_low_quality_entities_v2';
1735
1735
1736
1736
const migrationRecord = db.prepare('SELECT * FROM migrations WHERE id = ?').get(MIGRATION_ID) as { status: string } | undefined;
1737
1737
if (migrationRecord && migrationRecord.status === 'complete') return;
···
1798
1798
}
1799
1799
}
1800
1800
1801
-
// 5. Low-confidence addresses (regex-extracted places below new threshold)
1802
-
if (entityType === 'place' && confidence < 0.7 && (meta.extractor === 'regex')) {
1801
+
// 5. All regex-extracted places — the address regex is too noisy.
1802
+
// Going forward, regex places get confidence 0.6 which is below the 0.7 threshold,
1803
+
// so none will be stored. Clean up all existing ones.
1804
+
if (entityType === 'place' && meta.extractor === 'regex') {
1803
1805
removeEntity(entity.id);
1804
1806
continue;
1805
1807
}