tasks/task_medium.py

"""
tasks/task_medium.py -- Medium task: 3 variables, moderate noise, random domain.

The agent must discover multiple causal relationships with moderate noise
and a standard budget.

Grader returns 0.0-1.0 based on:
  - Hypothesis accuracy (50%)
  - Precision bonus (15%)
  - Efficiency bonus (15%)
  - Calibration (20%)
"""

from __future__ import annotations

from typing import Any

TASK_MEDIUM = {
    "id": "medium",
    "name": "Medium -- Multi-Edge Discovery",
    "description": (
        "Discover causal relationships among three variables. "
        "Moderate noise (sigma=0.20), standard budget (10 steps)."
    ),
    "difficulty": "medium",
    "reset_kwargs": {
        "noise_level": "medium",
        "seed": 123,
    },
}


def grade_medium(episode_result: dict[str, Any]) -> float:
    """
    Grade a medium-task episode. Returns a score in [0.0, 1.0].
    """
    accuracy = episode_result.get("accuracy_score", 0.0)
    precision = episode_result.get("precision_bonus", 0.0)
    efficiency = episode_result.get("efficiency_bonus", 0.0)
    calibration = episode_result.get("calibration_score", 0.0)

    raw = (
        0.50 * min(accuracy, 1.0)
        + 0.15 * min(precision / 0.10, 1.0)
        + 0.15 * min(efficiency / 0.15, 1.0)
        + 0.20 * min(calibration / 0.20, 1.0)
    )

    return round(max(0.0, min(1.0, raw)), 4)