"""
Rotary Stability rubric scorer — pure function, no model calls.

FMS Rotary Stability Criteria:
- Score 3: unilateral (same-side) arm/leg extension with trunk stable,
           elbow/knee touch performed smoothly.
- Score 2: contralateral (opposite) arm/leg extension performed with trunk stable.
- Score 1: inability to maintain trunk stability during contralateral pattern.
- Score 0: PAIN (spinal flexion clearing test) — never auto-scored.
"""
from __future__ import annotations

from formscout.types import BiomechFeatures, ScoreResult


def score_rotary_stability(features: BiomechFeatures) -> ScoreResult:
    """Pure rubric scorer for rotary stability."""
    angles = features.angles
    alignments = features.alignments

    has_data = "trunk_stability_std_px" in angles or "shoulder_level_diff_px" in angles
    if not has_data:
        return ScoreResult(
            score=1, rationale="Insufficient data: trunk stability not measurable",
            confidence=0.3, notes="missing key measurements",
        )

    trunk_stable = alignments.get("trunk_stable", False)
    shoulders_level = alignments.get("shoulders_level", False)
    hips_level = alignments.get("hips_level", False)

    rationale_parts = []

    # Without video classification of ipsi vs contra, assume contralateral (safer)
    if trunk_stable and shoulders_level and hips_level:
        score = 2  # Assume contralateral unless classifier says ipsilateral
        rationale_parts.append("Trunk stable during extension, shoulders and hips level")
        rationale_parts.append("scored as contralateral pattern (default)")
    elif trunk_stable or (shoulders_level and hips_level):
        score = 2
        if not trunk_stable:
            rationale_parts.append("minor trunk instability")
        rationale_parts.insert(0, "Contralateral pattern with minor compensation")
    else:
        score = 1
        std = angles.get("trunk_stability_std_px", 0)
        rationale_parts.append(f"Trunk instability detected (std {std:.1f}px)")
        if not shoulders_level:
            rationale_parts.append("shoulder asymmetry during extension")

    confidence = features.confidence * 0.75  # Lower confidence — hard to assess from 2D

    return ScoreResult(
        score=score, rationale="; ".join(rationale_parts),
        confidence=confidence, notes="ipsi/contra distinction requires VLM classifier",
    )