from typing import Callable, NamedTuple


class _Rule(NamedTuple):
    name: str
    decision: str
    matches: Callable[[dict, dict], bool]
    confidence: Callable[[dict, dict], float]


# Floor ESCALATE at 0.5 — we'd rather over-escalate than miss something urgent.
# Other rules scale proportionally with sentiment confidence.
RULES: list[_Rule] = [
    _Rule(
        name="negative_high_urgency",
        decision="ESCALATE",
        matches=lambda s, sig: s["label"] == "NEGATIVE" and sig["urgency"] == "HIGH",
        confidence=lambda s, sig: round(min(1.0, 0.5 + s["confidence"] * 0.5), 3),
    ),
    _Rule(
        name="negative_complaint",
        decision="FOLLOW_UP",
        matches=lambda s, sig: s["label"] == "NEGATIVE" and sig["intent"] == "COMPLAINT",
        confidence=lambda s, sig: round(min(1.0, 0.4 + s["confidence"] * 0.4), 3),
    ),
    _Rule(
        name="positive_praise",
        decision="LOG_FEEDBACK",
        matches=lambda s, sig: s["label"] == "POSITIVE" and sig["intent"] == "PRAISE",
        confidence=lambda s, sig: round(min(1.0, 0.3 + s["confidence"] * 0.6), 3),
    ),
    _Rule(
        name="question_any_sentiment",
        decision="AUTO_RESOLVE",
        matches=lambda s, sig: sig["intent"] == "QUESTION",
        confidence=lambda s, sig: 0.75,
    ),
]

_FALLBACK_DECISION = "FOLLOW_UP"
_FALLBACK_RULE = "fallback"


def _fallback_confidence(sentiment: dict) -> float:
    # Low ceiling (0.4) signals that no strong rule fired.
    return round(min(0.4, 0.2 + sentiment["confidence"] * 0.2), 3)


def decide(sentiment: dict, signals: dict) -> dict:
    for rule in RULES:
        if rule.matches(sentiment, signals):
            return {
                "decision": rule.decision,
                "confidence": rule.confidence(sentiment, signals),
                "signals_used": {
                    "label": sentiment["label"],
                    "sentiment_confidence": sentiment["confidence"],
                    "urgency": signals["urgency"],
                    "intent": signals["intent"],
                    "rule_matched": rule.name,
                },
            }

    return {
        "decision": _FALLBACK_DECISION,
        "confidence": _fallback_confidence(sentiment),
        "signals_used": {
            "label": sentiment["label"],
            "sentiment_confidence": sentiment["confidence"],
            "urgency": signals["urgency"],
            "intent": signals["intent"],
            "rule_matched": _FALLBACK_RULE,
        },
    }