1. Reddit app credentials (PRAW client_id + client_secret)
2. Daily cron: pull new posts + top comments from tracked subreddits, upsert into reddit_posts table
3. Entity-linker: match post body mentions → products.upc / brands.name / producers.name. Start with brand-name match, extend with strain-name lexicon.
4. Sentiment scoring: off-the-shelf (VADER) for v0, fine-tuned model for v1. Output: post_entity_sentiment(post_id, entity_type, entity_id, score, confidence).
5. Aggregator: roll up to daily / weekly by entity, hook into the same window pattern as the rest of the app.

Once the table exists, this page connects the same way /trends and /opportunities do — via a sentimentWindow(w) aggregator in lib/cached.ts.