Public Access
Captures brainstorm decisions for the meal-planner algorithm and the minimum recipe-engine surface needed to feed it: - 6 hard constraints (blocklist, never_suggest, recency N=4, calories +/-20%, time <=45min, cost <=$30) - Top-K=20 set enumeration with diversity penalty for protein/cuisine - Ingredient<->grocery_item matching as a cacheable layer (rapidfuzz + manual override) rather than per-run fuzzy work - Thin phase 4: recipe CRUD, ingredient CRUD, resolve-ingredient assist, manual match override, match job, 30-recipe seed - Recipe ingestion source pros/cons (TheMealDB + Spoonacular + manual) documented; decision deferred until phase 4 + phase 9 work end-to-end Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>