{"data":{"kind":"file","path":"README.md","version_id":"vm6k2z1jken1mkl02q9lotxt","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2151,"modified_at":"2026-08-08T01:31:28.699000","content_hash":"0a5980e3528184b0d21775191b9ebba5687aaf38f49c45b166f53f3348922375"},"entries":[],"content":"# Recipe & Cooking Planner\n\nEvaluate LLM capabilities in culinary planning, recipe generation, and food science reasoning.\n\n## What it tests\n\n| Metric | Weight | Description |\n|--------|--------|-------------|\n| Recipe Completeness | 1.0 | Ingredients, steps, quantities, timing, servings |\n| Step Logic | 0.8 | Actionable verbs, temperatures, logical transitions |\n| Ingredient Coverage | 1.0 | Uses all required ingredients from the task |\n| Dietary Adaptation | 0.7 | Substitution quality with reasoning for restrictions |\n| Cooking Science | 0.6 | Food science explanations (Maillard, fermentation, etc.) |\n| Meal Plan Quality | 0.7 | Structure, variety, macros, shopping lists |\n| Response Structure | 0.5 | Headers, lists, reasonable length, organization |\n\n## Task types\n\n1. **Full recipe generation** — Create complete dishes from scratch (Pad Thai, Sourdough, etc.)\n2. **Ingredient-constrained cooking** — Work with what the user has in the fridge\n3. **Meal planning** — Multi-day plans with nutritional targets (vegetarian muscle building)\n4. **Dietary adaptation** — Convert classics to gluten-free, dairy-free, nut-free\n5. **Allergy-safe cooking** — Work around specific allergies with alternatives\n6. **Food science queries** — Explain the why behind techniques\n7. **Kid-friendly planning** — Lunchbox meals with constraints (nut-free, <20 min prep)\n8. **Homemade from scratch** — Complex processes with starter cultures and troubleshooting\n\n## Usage\n\n```bash\n# Run eval\nprime eval run OWNER/recipe-cooking-planner -m meta-llama/Llama-3.2-1B-Instruct --plain\n\n# Run with specific args\nprime eval run OWNER/recipe-cooking-planner -m Qwen/Qwen3-0.5B --env-args '{}' --plain\n```\n\n## Design\n\n- **7 rubric functions** covering structure, logic, science, and dietary handling\n- **8 diverse tasks** spanning different cuisines, diets, and complexity levels\n- **JSON-structured ground truth** with required components, time ranges, and technique lists\n- **ThinkParser** to encourage reasoning before answering\n- Anti-gaming: requires actual ingredient names, cooking verbs, and transitional phrases — not just formatting tricks\n","encoding":"utf-8","truncated":false,"total_bytes":2151},"status":null}