# Rule-based triage classifier for support inquiries.
#
# Deterministic keyword scorer: every category carries a list of case-insensitive
# substrings (single words and multi-word phrases).  classify() returns the
# category with the most matching keywords; ties are resolved by a fixed priority
# order.  Texts with zero keyword hits fall back to "other".
#
# This is a dictionary-exact router (triage lane).  Real-world accuracy is
# measured, not assumed: the gold set in gold.json and the assertion in
# test_triage.py pin the expected behaviour, and run_audit.py reports precision /
# recall / f1 per category.

CATEGORY_ORDER = [
    "account",
    "billing",
    "technical_support",
    "feature_request",
    "other",
]

CATEGORY_META = {
    "account": "Account access & identity",
    "billing": "Billing & payments",
    "technical_support": "Technical support",
    "feature_request": "Feature request",
    "other": "Other / general",
}

# Multi-word keywords are matched as substrings too, so "log in" matches
# "cannot log in" (which is purely account-side speech).
KEYWORDS = {
    "account": [
        "log in",
        "password",
        "2fa",
        "sign up",
        "signup",
        "cannot log in",
        "verification code",
        "not received my code",
    ],
    "billing": [
        "invoice",
        "charged",
        "refund",
        "subscription",
        "credit card",
        "payment method",
        "amount due",
        "billing statement",
    ],
    "technical_support": [
        "crash",
        "500",
        "slow",
        "deploy",
        "cannot connect",
        "connection refused",
        "database unreachable",
        "page not found",
        "timeout",
        "install",
        "uptime check",
        "api error",
        "experienced an error",
    ],
    "feature_request": [
        "feature",
        "suggest",
        "enhancement",
        "request",
        "would like",
        "new feature",
        "add ability",
    ],
}


def classify(text):
    """Return (category, scores_dict) for a single inquiry string."""
    low = text.lower()
    scores = {"other": 0}
    scores.update({cat: sum(1 for k in kws if k in low) for cat, kws in KEYWORDS.items()})
    best = max(scores.values())
    if best == 0:
        return "other", scores
    winners = [c for c in CATEGORY_ORDER if scores[c] == best]
    return winners[0], scores


def batch_classify(texts):
    """Classify a list of texts; result mirrors (text, category)."""
    return [classify(t) for t in texts]