scorer.py

from dataclasses import dataclass
from typing import Dict, Any, List
import re

REQ = [
    "decoherence_onset_timestamp",
    "coherence_drop_delta",
    "affected_modalities",
    "narrative_conflict_flag",
    "onset_confidence",
    "early_warning_score",
]

@dataclass
class ScoreResult:
    score: float
    details: Dict[str, Any]

def _time_ok(p: str):
    # accepts t=6.2s or 6.2s
    m = re.search(r"decoherence_onset_timestamp\s*[:=]\s*(t\s*=\s*)?([0-9]+(\.[0-9]+)?)\s*s", p)
    if not m:
        return None
    return float(m.group(2))

def _f(p: str, key: str):
    m = re.search(rf"{key}\s*[:=]\s*(0\.\d+|1\.0)\b", p)
    return float(m.group(1)) if m else None

def _i(p: str, key: str):
    m = re.search(rf"{key}\s*[:=]\s*(\d+)\b", p)
    return int(m.group(1)) if m else None

def score(sample: Dict[str, Any], prediction: str) -> ScoreResult:
    p = (prediction or "").lower()
    words_ok = len(p.split()) <= 900

    hits = sum(1 for k in REQ if k in p)

    t = _time_ok(p)
    delta = _f(p, "coherence_drop_delta")
    conf = _f(p, "onset_confidence")
    warn = _f(p, "early_warning_score")
    flag = _i(p, "narrative_conflict_flag")

    numeric_ok = int(
        t is not None and 0.0 <= t <= 120.0 and
        delta is not None and 0.0 <= delta <= 1.0 and
        conf is not None and 0.0 <= conf <= 1.0 and
        warn is not None and 0.0 <= warn <= 1.0 and
        flag is not None and flag in [0, 1]
    )

    mods_ok = int("affected_modalities" in p and len(p) > 70)

    # optional sanity: if delta is high, warning should tend high
    sanity = 0
    if delta is not None and warn is not None:
        sanity = int(warn + 0.15 >= delta)

    raw = (
        0.15 * int(words_ok) +
        0.45 * (hits / len(REQ)) +
        0.20 * numeric_ok +
        0.10 * mods_ok +
        0.10 * sanity
    )

    return ScoreResult(score=min(1.0, raw), details={"id": sample.get("id"), "hits": hits, "sanity": sanity})

def aggregate(results: List[ScoreResult]) -> Dict[str, Any]:
    if not results:
        return {"mean": 0.0, "n": 0}
    return {"mean": sum(r.score for r in results)/len(results), "n": len(results)}