"""Conservative confidence defaults for underrepresented classes.

When calibration data is insufficient for a specific document type or
event class, returns conservative values (0.3-0.5) and marks the result
as under-calibrated per Requirement 10.7.
"""

from __future__ import annotations

import logging

from services.intelligence_pipeline_v3.confidence.models import ConfidenceResult

logger = logging.getLogger(__name__)

# Conservative default probabilities by document type.
# These are intentionally low (0.3-0.5) to avoid overconfidence
# when insufficient calibration data exists.
_DOCUMENT_TYPE_DEFAULTS: dict[str, float] = {
    "news": 0.45,
    "filing": 0.40,
    "transcript": 0.40,
    "press_release": 0.45,
    "macro_event": 0.35,
    "unknown": 0.30,
}

# Conservative default probabilities by event class.
# More complex or rare event types get lower defaults.
_EVENT_CLASS_DEFAULTS: dict[str, float] = {
    "earnings_beat": 0.50,
    "earnings_miss": 0.50,
    "guidance_raise": 0.45,
    "guidance_cut": 0.45,
    "merger_acquisition": 0.40,
    "product_launch": 0.45,
    "regulatory_action": 0.40,
    "management_change": 0.45,
    "legal_proceeding": 0.40,
    "supply_chain": 0.35,
    "rating_change": 0.45,
    "dividend_change": 0.45,
    "buyback": 0.45,
    "macro_policy": 0.35,
    "geopolitical": 0.30,
    "sector_rotation": 0.35,
    "unknown": 0.30,
}

# Features used when returning conservative defaults
_DEFAULT_FEATURES_USED = [
    "document_type_prior",
    "event_class_prior",
]


def get_default_confidence(
    document_type: str,
    event_class: str,
) -> ConfidenceResult:
    """Return a conservative confidence result for underrepresented classes.

    Used when calibration data is insufficient for the given document type
    and event class combination. Returns conservative probabilities (0.3-0.5)
    and marks the result as under-calibrated.

    Parameters
    ----------
    document_type
        The document type (news, filing, transcript, etc.).
    event_class
        The classified event type (earnings_beat, merger_acquisition, etc.).

    Returns
    -------
    ConfidenceResult
        A conservative confidence result with under_calibrated=True.
    """
    doc_default = _DOCUMENT_TYPE_DEFAULTS.get(
        document_type, _DOCUMENT_TYPE_DEFAULTS["unknown"]
    )
    event_default = _EVENT_CLASS_DEFAULTS.get(
        event_class, _EVENT_CLASS_DEFAULTS["unknown"]
    )

    # Take the minimum of document and event defaults for extra conservatism
    probability = min(doc_default, event_default)

    logger.debug(
        "Using conservative default confidence: doc_type=%s (%.2f), event=%s (%.2f) -> %.2f",
        document_type,
        doc_default,
        event_class,
        event_default,
        probability,
    )

    return ConfidenceResult(
        probability=probability,
        features_used=_DEFAULT_FEATURES_USED,
        is_calibrated=False,
        under_calibrated=True,
        calibration_version="conservative-default-v1",
    )


def is_underrepresented(
    document_type: str,
    event_class: str,
    min_samples: int = 30,
    known_counts: dict[tuple[str, str], int] | None = None,
) -> bool:
    """Check if a document_type + event_class combination is underrepresented.

    Parameters
    ----------
    document_type
        The document type.
    event_class
        The event class.
    min_samples
        Minimum number of calibration samples to consider a class well-represented.
    known_counts
        Optional mapping of (doc_type, event_class) -> sample count.
        If None, treats any unknown combination as underrepresented.

    Returns
    -------
    bool
        True if the class has insufficient calibration data.
    """
    if known_counts is None:
        # Without explicit counts, use heuristic: unknown types are underrepresented
        if document_type not in _DOCUMENT_TYPE_DEFAULTS:
            return True
        if event_class not in _EVENT_CLASS_DEFAULTS:
            return True
        return False

    key = (document_type, event_class)
    count = known_counts.get(key, 0)
    return count < min_samples