scorer.py

import csv
import json
import sys
from typing import Any, Dict, List


LABEL_COL = "label_mof_boundary"
POSITIVE_LABEL = 1

PRIMARY_METRIC = "recall_correct_control_sequence_selection"
SECONDARY_METRIC = "false_effective_control_rate"

CONTROL_ALIGNMENT_THRESHOLD = 0.60
INTERVENTION_ALIGNMENT_THRESHOLD = 0.60
RECOVERY_CONSISTENCY_THRESHOLD = 0.60
HIGH_UNCERTAINTY_THRESHOLD = 0.25
NARROW_WINDOW_THRESHOLD = 0.30
OSCILLATION_THRESHOLD = 0.25
STABLE_MARGIN_THRESHOLD = 0.60
TRAJECTORY_SHIFT_THRESHOLD = -0.10


def _to_float(value: Any, default: float = 0.0) -> float:
    try:
        return float(value)
    except (TypeError, ValueError):
        return default


def _to_int(value: Any, default: int = 0) -> int:
    try:
        return int(float(value))
    except (TypeError, ValueError):
        return default


def _safe_div(numerator: float, denominator: float) -> float:
    return numerator / denominator if denominator else 0.0


def _read_csv(path: str) -> List[Dict[str, str]]:
    with open(path, "r", encoding="utf-8", newline="") as f:
        return list(csv.DictReader(f))


def _normalize_string(value: Any) -> str:
    if value is None:
        return ""
    return str(value).strip().lower()


def _normalize_label(row: Dict[str, str]) -> int:
    if LABEL_COL in row:
        return _to_int(row[LABEL_COL])

    for key, value in row.items():
        if key.startswith("label_"):
            return _to_int(value)

    raise ValueError(
        f"No label column found in row. Expected '{LABEL_COL}' or a column starting with 'label_'."
    )


def _confusion(y_true: List[int], y_pred: List[int]) -> Dict[str, int]:
    tp = fp = tn = fn = 0
    for yt, yp in zip(y_true, y_pred):
        if yt == 1 and yp == 1:
            tp += 1
        elif yt == 0 and yp == 1:
            fp += 1
        elif yt == 0 and yp == 0:
            tn += 1
        elif yt == 1 and yp == 0:
            fn += 1
    return {"tp": tp, "fp": fp, "tn": tn, "fn": fn}


def _accuracy(y_true: List[int], y_pred: List[int]) -> float:
    return _safe_div(sum(1 for a, b in zip(y_true, y_pred) if a == b), len(y_true))


def _precision(cm: Dict[str, int]) -> float:
    return _safe_div(cm["tp"], cm["tp"] + cm["fp"])


def _recall(cm: Dict[str, int]) -> float:
    return _safe_div(cm["tp"], cm["tp"] + cm["fn"])


def _f1(precision: float, recall: float) -> float:
    return _safe_div(2 * precision * recall, precision + recall)


def _mean_abs_error(true_vals: List[float], pred_vals: List[float]) -> float:
    if not true_vals:
        return 0.0
    return sum(abs(a - b) for a, b in zip(true_vals, pred_vals)) / len(true_vals)


def _rate(mask: List[bool], condition: List[bool]) -> float:
    idx = [i for i, flag in enumerate(mask) if flag]
    if not idx:
        return 0.0
    return sum(1 for i in idx if condition[i]) / len(idx)


def _string_accuracy(true_vals: List[str], pred_vals: List[str]) -> float:
    if not true_vals:
        return 0.0
    return _safe_div(
        sum(
            1 for a, b in zip(true_vals, pred_vals)
            if _normalize_string(a) == _normalize_string(b)
        ),
        len(true_vals),
    )


def _threshold_accuracy(
    gold: List[Dict[str, str]],
    pred: List[Dict[str, str]],
    field: str,
    threshold: float,
) -> float:
    true_flags = [_to_float(row.get(field, 0.0)) >= threshold for row in gold]
    pred_flags = [_to_float(row.get(field, 0.0)) >= threshold for row in pred]
    return _safe_div(
        sum(1 for a, b in zip(true_flags, pred_flags) if a == b),
        len(true_flags),
    )


def _validate_inputs(gold: List[Dict[str, str]], pred: List[Dict[str, str]]) -> None:
    if len(gold) != len(pred):
        raise ValueError("Solution and submission row counts do not match.")

    if not gold:
        raise ValueError("Solution file is empty.")

    if not pred:
        raise ValueError("Submission file is empty.")

    gold_has_id = all("scenario_id" in row for row in gold)
    pred_has_id = all("scenario_id" in row for row in pred)

    if gold_has_id and pred_has_id:
        for i, (g_row, p_row) in enumerate(zip(gold, pred)):
            if _normalize_string(g_row.get("scenario_id")) != _normalize_string(p_row.get("scenario_id")):
                raise ValueError(
                    f"Scenario mismatch at row {i + 1}: "
                    f"{g_row.get('scenario_id')} != {p_row.get('scenario_id')}"
                )


def score(solution_file: str, submission_file: str) -> Dict[str, Any]:
    gold = _read_csv(solution_file)
    pred = _read_csv(submission_file)

    _validate_inputs(gold, pred)

    y_true = [_normalize_label(row) for row in gold]
    y_pred = [_normalize_label(row) for row in pred]

    cm = _confusion(y_true, y_pred)
    accuracy = _accuracy(y_true, y_pred)
    precision = _precision(cm)
    recall = _recall(cm)
    f1 = _f1(precision, recall)

    # Primary v1.0 success condition:
    # A row counts as a true control success only when the system truly stabilized,
    # the trajectory improved meaningfully, the intervention was aligned,
    # the control sequence itself was aligned, and recovery remained coherent.
    primary_true = [
        _to_int(row.get("stabilization_success", 0)) == 1
        and _to_float(row.get("trajectory_shift", 0.0)) < TRAJECTORY_SHIFT_THRESHOLD
        and _to_float(row.get("intervention_alignment_score", 0.0)) >= INTERVENTION_ALIGNMENT_THRESHOLD
        and _to_float(row.get("control_sequence_alignment_score", 0.0)) >= CONTROL_ALIGNMENT_THRESHOLD
        and _to_float(row.get("recovery_consistency_score", 0.0)) >= RECOVERY_CONSISTENCY_THRESHOLD
        for row in gold
    ]
    primary_pred = [pred_label == POSITIVE_LABEL for pred_label in y_pred]

    recall_correct_control_sequence_selection = _rate(primary_true, primary_pred)

    false_effective_control_rate = _rate(
        primary_pred,
        [not flag for flag in primary_true]
    )

    primary_intervention_path_accuracy = _string_accuracy(
        [row.get("primary_intervention_path", "") for row in gold],
        [row.get("primary_intervention_path", "") for row in pred],
    )

    secondary_intervention_path_accuracy = _string_accuracy(
        [row.get("secondary_intervention_path", "") for row in gold],
        [row.get("secondary_intervention_path", "") for row in pred],
    )

    control_sequence_alignment_accuracy = _threshold_accuracy(
        gold, pred, "control_sequence_alignment_score", CONTROL_ALIGNMENT_THRESHOLD
    )
    feedback_response_accuracy = _threshold_accuracy(
        gold, pred, "feedback_response_score", CONTROL_ALIGNMENT_THRESHOLD
    )
    intervention_timing_accuracy = _threshold_accuracy(
        gold, pred, "intervention_timing_score", CONTROL_ALIGNMENT_THRESHOLD
    )

    missed_rescues = [
        primary_true[i] and not primary_pred[i]
        for i in range(len(primary_true))
    ]

    # Miss-conditioned diagnostics resolve to 0.0 by design when there are zero missed rescues.
    # Of missed rescues, what fraction occurred under high intervention uncertainty.
    high_uncertainty_control_miss_rate = _rate(
        missed_rescues,
        [_to_float(row.get("intervention_uncertainty", 0.0)) >= HIGH_UNCERTAINTY_THRESHOLD for row in gold]
    )

    # Of missed rescues, what fraction occurred in narrow rescue windows.
    narrow_window_control_miss_rate = _rate(
        missed_rescues,
        [_to_float(row.get("rescue_window_width", 0.0)) <= NARROW_WINDOW_THRESHOLD for row in gold]
    )

    adaptation_latency_error = _mean_abs_error(
        [_to_float(row.get("adaptation_latency", 0.0)) for row in gold],
        [_to_float(row.get("adaptation_latency", 0.0)) for row in pred],
    )

    control_stability_error = _mean_abs_error(
        [_to_float(row.get("control_stability_margin", 0.0)) for row in gold],
        [_to_float(row.get("control_stability_margin", 0.0)) for row in pred],
    )

    recovery_consistency_error = _mean_abs_error(
        [_to_float(row.get("recovery_consistency_score", 0.0)) for row in gold],
        [_to_float(row.get("recovery_consistency_score", 0.0)) for row in pred],
    )

    control_horizon_error = _mean_abs_error(
        [_to_float(row.get("control_horizon", 0.0)) for row in gold],
        [_to_float(row.get("control_horizon", 0.0)) for row in pred],
    )

    recalibration_overuse_rate = _rate(
        [True] * len(gold),
        [
            _to_int(pred[i].get("control_recalibration_count", 0)) >
            _to_int(gold[i].get("control_recalibration_count", 0))
            and _to_float(gold[i].get("control_stability_margin", 0.0)) >= STABLE_MARGIN_THRESHOLD
            for i in range(len(gold))
        ]
    )

    controller_oscillation_misread_rate = _rate(
        [True] * len(gold),
        [
            (_to_float(gold[i].get("controller_oscillation_score", 0.0)) >= OSCILLATION_THRESHOLD)
            != (_to_float(pred[i].get("controller_oscillation_score", 0.0)) >= OSCILLATION_THRESHOLD)
            for i in range(len(gold))
        ]
    )

    terminal_pathway_state_accuracy = _string_accuracy(
        [row.get("terminal_pathway_state", "") for row in gold],
        [row.get("terminal_pathway_state", "") for row in pred],
    )

    positive_class_count = sum(y_true)
    negative_class_count = len(y_true) - positive_class_count
    primary_true_count = sum(primary_true)
    primary_pred_count = sum(primary_pred)
    missed_rescue_count = sum(missed_rescues)

    result = {
        "primary_metric": PRIMARY_METRIC,
        "secondary_metric": SECONDARY_METRIC,
        "binary_metrics": {
            "accuracy": round(accuracy, 6),
            "precision": round(precision, 6),
            "recall": round(recall, 6),
            "f1": round(f1, 6),
            "confusion_matrix": cm
        },
        "diagnostics": {
            PRIMARY_METRIC: round(recall_correct_control_sequence_selection, 6),
            SECONDARY_METRIC: round(false_effective_control_rate, 6),
            "primary_intervention_path_accuracy": round(primary_intervention_path_accuracy, 6),
            "secondary_intervention_path_accuracy": round(secondary_intervention_path_accuracy, 6),
            "control_sequence_alignment_accuracy": round(control_sequence_alignment_accuracy, 6),
            "control_horizon_error": round(control_horizon_error, 6),
            "feedback_response_accuracy": round(feedback_response_accuracy, 6),
            "intervention_timing_accuracy": round(intervention_timing_accuracy, 6),
            "high_uncertainty_control_miss_rate": round(high_uncertainty_control_miss_rate, 6),
            "narrow_window_control_miss_rate": round(narrow_window_control_miss_rate, 6),
            "adaptation_latency_error": round(adaptation_latency_error, 6),
            "control_stability_error": round(control_stability_error, 6),
            "recovery_consistency_error": round(recovery_consistency_error, 6),
            "recalibration_overuse_rate": round(recalibration_overuse_rate, 6),
            "controller_oscillation_misread_rate": round(controller_oscillation_misread_rate, 6),
            "terminal_pathway_state_accuracy": round(terminal_pathway_state_accuracy, 6)
        },
        "support": {
            "num_rows": len(gold),
            "primary_true_count": primary_true_count,
            "primary_pred_count": primary_pred_count,
            "missed_rescue_count": missed_rescue_count,
            "positive_class_count": positive_class_count,
            "negative_class_count": negative_class_count
        },
        "class_balance": {
            "positive_rate": round(_safe_div(positive_class_count, len(y_true)), 6),
            "negative_rate": round(_safe_div(negative_class_count, len(y_true)), 6)
        },
        "thresholds": {
            "control_alignment_threshold": CONTROL_ALIGNMENT_THRESHOLD,
            "intervention_alignment_threshold": INTERVENTION_ALIGNMENT_THRESHOLD,
            "recovery_consistency_threshold": RECOVERY_CONSISTENCY_THRESHOLD,
            "high_uncertainty_threshold": HIGH_UNCERTAINTY_THRESHOLD,
            "narrow_window_threshold": NARROW_WINDOW_THRESHOLD,
            "oscillation_threshold": OSCILLATION_THRESHOLD,
            "stable_margin_threshold": STABLE_MARGIN_THRESHOLD,
            "trajectory_shift_threshold": TRAJECTORY_SHIFT_THRESHOLD
        }
    }

    return result


if __name__ == "__main__":
    if len(sys.argv) != 3:
        raise SystemExit("Usage: python scorer.py <solution_file> <submission_file>")

    scores = score(sys.argv[1], sys.argv[2])
    print(json.dumps(scores, indent=2))