NEW MODULE app/catalog.py:
- refresh_catalog(cats, sources, per_brand, delay) — runs parsers for seed brand+category pairs
- list_catalog(cat, tier, brand) — reads from Sheets
- list_for_ai(cats, tiers) — compact text for AI prompt context
- SEED_BRANDS_BY_TIER + CATEGORY_QUERIES — 22 brands × 8 cats = 176 combos
- Saves top-2 relevant results per (brand × cat), filters by brand presence in title
- Dedup by title hash within (cat, brand) bucket
SHEETS:
- ensure_sheet(name, headers) — auto-creates Catalog tab on first refresh
- Schema: id, category, brand, tier, model_name, search_query, price_min/max, image_url, source, url, last_seen_at
ENDPOINTS:
- POST /api/catalog/refresh?cat=X&per_brand=N — manual refresh (1 cat ~2-5 min)
- GET /api/catalog/list?cat=&tier=&brand= — read with filters
- GET /api/catalog/preview_ai?cats=fridge — debug what AI receives
AI PROMPT:
- Rule #0: if catalog passed in user prompt — MUST select only from there
- _build_catalog_context: filters by checklist.budget_preset → tier subset
(luxe→premium, premium→premium+middle, middle→middle, budget→middle+budget)
_handle_podbor:
- Loads catalog subset, appends to user_prompt as 'ДОСТУПНЫЙ КАТАЛОГ МОДЕЛЕЙ'
- AI 'выбирай ТОЛЬКО из этого списка' rule reinforced
NEXT: trigger refresh manually for 1 category (~3 min), then real podbor test
to verify AI uses catalog models instead of hallucinating SKUs
