- hackernews: use extract_core_subject instead of raw topic, add
points>5 filter and restrictSearchableAttributes=title to reduce
noise from URL-match and low-signal posts
- youtube: add --dateafter parameter to yt-dlp for server-side date
filtering (Python soft filter still handles fallback)
- reddit: skip opinion/review query variant for how_to/comparison
queries where it adds noise
- bird_x: add OR-group retry with compound terms before falling back
to word-dropping (uses X OR operator for multi-concept queries)
- query.py: add detect_query_type() and extract_compound_terms()
Consolidate duplicated _extract_core_subject() (7 copies across bird_x,
reddit, youtube_yt, tiktok, instagram, bluesky, scrapecreators_x) into
query.extract_core_subject() with parameterized noise set, max_words,
and suffix stripping.
Consolidate duplicated _tokenize/_compute_relevance/STOPWORDS/SYNONYMS
(4 copies across youtube_yt, tiktok, instagram, scrapecreators_x) into
relevance.token_overlap_relevance() with hashtag-aware matching.
Integration into per-module imports follows in next commits.