{"task": {"agent_timeout": 1800, "task": "llmsr-bench-chem_react-crk17", "verifier_timeout": 600, "instruction": "# Scientific Equation Discovery: CRK17\n\n## Problem Description\n\nYour task is to discover a mathematical equation that describes the relationship between the given variables using the provided experimental data.\n\n## Variables\n\n- `dA_dt`: Rate of change of concentration in chemistry reaction kinetics (type: output)\n- `t`: Time (type: input)\n- `A`: Concentration at time t (type: input)\n\n## Task\n\n1. Analyze the training data\n2. Discover the equation relating inputs to output\n3. Write the equation to `/logs/agent/discovered_equation.txt`\n\n## Data Files\n\n**Training data (use this to discover equation):**\n- `/data/train_data.csv`\n\n**Test data (used for scoring - don't look at this!):**\n- `/data/test_data.csv`\n\n**Optional OOD test (if exists):**\n- `/data/ood_test_data.csv`\n\n> **Note:** In all CSV files, the **last column** is the output variable and all earlier columns are inputs.\n\n## Equation Format\n\n### Basic Syntax\n```python\ndA_dt = your_expression_here\n```\n\n### Available Operations\n\n- **Arithmetic:** `+`, `-`, `*`, `/`, `**` (power)\n- **Functions:** `sqrt()`, `exp()`, `log()`, `sin()`, `cos()`, `tan()`, `abs()`\n- **Constants:** Use variable names from CSV headers exactly\n\n### Free Parameters\n\nYou may introduce **scalar constants** (e.g., `c0`, `c1`, `k`, `alpha`) that are **not** CSV columns. These will be **automatically fitted** using BFGS optimization during evaluation.\n\n## Evaluation Metrics\n\nYour equation will be scored using:\n\n- **R\u00b2** (coefficient of determination): **This is your reward!**\n  - 1.0 = Perfect prediction\n  - 0.9+ = Excellent\n  - 0.8+ = Good\n  - 0.6+ = Moderate\n  - <0.6 = Poor\n\n- **MSE** (mean squared error): Logged for analysis\n- **NMSE** (normalized MSE): Logged for analysis\n\n## Output Format\n\nWrite **exactly one line** to `/logs/agent/discovered_equation.txt`:\n```python\ndA_dt = your_expression_here\n```\n", "memory": "", "runnable": false, "difficulty": "medium", "language": "", "cpus": "", "instruction_truncated": false, "category": "scientific-equation-discovery", "compose": false, "has_solution": true, "oracle": null, "docker_image": "", "taskset": "llmsr-bench", "tags": ["symbolic-regression", "equation-discovery", "scientific-computing"]}, "runs": []}