"""Lexical scoring primitives shared by runtime gates and evaluation."""

import re
import unicodedata

_TOKEN_RE = re.compile(r"[a-z0-9]+")
HINDI_MONTHS = {
    "जनवरी": 1, "फरवरी": 2, "मार्च": 3, "अप्रैल": 4, "मई": 5, "जून": 6,
    "जुलाई": 7, "अगस्त": 8, "सितंबर": 9, "सितम्बर": 9, "अक्टूबर": 10,
    "नवंबर": 11, "नवम्बर": 11, "दिसंबर": 12, "दिसम्बर": 12,
}


def tokens(text: str) -> list[str]:
    """Lowercase alphanumeric tokens (ROUGE-style basic tokenization)."""
    return _TOKEN_RE.findall(text.lower())


def word_base(token: str) -> str:
    """Crude morphological base: strips a plural 's' so "rights" and
    "right" match (legal prose leans heavily on pluralized nouns)."""
    return token[:-1] if len(token) > 3 and token.endswith("s") else token


STOPWORDS = frozenset(
    [
        "a",
        "an",
        "and",
        "are",
        "as",
        "at",
        "be",
        "by",
        "for",
        "from",
        "has",
        "have",
        "in",
        "is",
        "it",
        "its",
        "of",
        "on",
        "or",
        "that",
        "the",
        "to",
        "under",
        "with",
        "within",
        "without",
        "shall",
        "may",
        "must",
    ]
)

_RESEARCH_STOP = STOPWORDS | frozenset(
    ["what", "how", "when", "why", "which", "who", "does", "do", "did", "was", "were", "can", "could", "should", "would", "find", "show", "give", "tell", "about", "explain", "summarize", "summarise", "please", "judgment", "judgement", "judgments", "judgements", "case", "cases", "court", "supreme", "india", "indian", "decision", "decisions", "verdict", "holding", "outcome", "reason", "reasons", "precedent", "precedents", "authority", "authorities", "final", "latest", "change", "changed", "happened", "order", "orders", "year", "date", "क्या", "कब", "क्यों", "कैसे", "कौन", "का", "की", "के", "को", "में", "से", "पर", "और", "है", "हैं", "था", "थी", "थे", "मुझे", "बताओ", "बताइए", "समझाओ", "फैसला", "फ़ैसला", "निर्णय", "अदालत", "न्यायालय"]
)


def research_terms(text: str) -> set[str]:
    """Issue terms for selecting excerpts; preserve Devanagari vowel marks.

    This does not rewrite the request, resolve a case or assert legal support.
    Existing English-only evaluation tokenization remains unchanged.
    """
    normalized = unicodedata.normalize("NFKC", text).casefold()
    normalized = normalized.translate(str.maketrans("०१२३४५६७८९", "0123456789"))
    return {
        word_base(term)
        for term in re.findall(r"[^\W_][\w\u0900-\u0903\u093a-\u094f\u0951-\u0957\u0962-\u0963]*", normalized)
        if (len(term) > 1 or term.isdigit()) and term not in _RESEARCH_STOP
    }