Blame for data.py

from __future__ import annotations

import json
import re
from pathlib import Path

FIXTURES_DIR = Path(__file__).parent / "fixtures"


def load_fixtures() -> dict:
    return {
        "policyholders": _load_json("policyholders.json"),
        "claims": _load_json("claims.json"),
        "claim_schema": _load_json("claim_schema.json"),
        "guidelines": _load_json("required_document_guideline.json"),
        "representatives": _load_json("representatives.json"),
        "consent_scenarios": _load_json("consent_scenarios.json"),
    }


def _load_json(filename: str):
    with open(FIXTURES_DIR / filename) as f:
        return json.load(f)


# ---------------------------------------------------------------------------
# PII matching
# ---------------------------------------------------------------------------

def _normalize_phone(raw: str) -> str:
    digits = re.sub(r"[^0-9]", "", raw)
    if len(digits) == 11 and digits.startswith("1"):
        digits = digits[1:]
    return digits


def _normalize_policy(raw: str) -> str:
    return re.sub(r"^pol[-\s]?", "", raw.strip(), flags=re.IGNORECASE).strip().upper()


def _normalize_date(raw: str) -> str | None:
    raw = raw.strip()
    iso_match = re.match(r"^(\d{4})-(\d{1,2})-(\d{1,2})$", raw)
    if iso_match:
        return f"{iso_match.group(1)}-{int(iso_match.group(2)):02d}-{int(iso_match.group(3)):02d}"
    us_match = re.match(r"^(\d{1,2})/(\d{1,2})/(\d{2,4})$", raw)
    if us_match:
        year = us_match.group(3)
        if len(year) == 2:
            century = "20" if int(year) <= 30 else "19"
            year = century + year
        return f"{year}-{int(us_match.group(1)):02d}-{int(us_match.group(2)):02d}"
    # "March 15, 1985", "15 March 1985", "Mar 15 1985", "March 15th, 1985" —
    # the model passes dates through in whatever form the caller spoke them.
    # "3 december 1989" used to return None, so a caller with correct details
    # failed verification until the model happened to re-record it as ISO.
    day, month_str, year = None, None, None
    verbal = re.match(r"^([a-zA-Z]+)\.?\s+(\d{1,2})(?:st|nd|rd|th)?,?\s+(\d{4})$", raw)
    if verbal:
        month_str, day, year = verbal.group(1), verbal.group(2), verbal.group(3)
    else:
        verbal = re.match(r"^(\d{1,2})(?:st|nd|rd|th)?\s+(?:of\s+)?([a-zA-Z]+)\.?,?\s+(\d{4})$", raw)
        if verbal:
            day, month_str, year = verbal.group(1), verbal.group(2), verbal.group(3)
    if month_str:
        month = _MONTHS.get(month_str.lower()) or _MONTHS.get(month_str.lower()[:3])
        if month:
            return f"{year}-{month}-{int(day):02d}"
    return None


_MONTHS = {
    "january": "01", "february": "02", "march": "03", "april": "04",
    "may": "05", "june": "06", "july": "07", "august": "08",
    "september": "09", "october": "10", "november": "11", "december": "12",
    "jan": "01", "feb": "02", "mar": "03", "apr": "04", "jun": "06", "jul": "07",
    "aug": "08", "sep": "09", "sept": "09", "oct": "10", "nov": "11", "dec": "12",
}


# The five identity factors the SOP accepts (SPEC.txt). A policy number is
# recorded because callers volunteer it and it is useful later, but it does not
# count toward the three — it is printed on every piece of mail.
VERIFICATION_FIELDS = ("full_name", "dob", "phone", "email", "id_last4")
PII_FIELDS = VERIFICATION_FIELDS + ("policy_number",)


def verification_field_count(collected_pii: dict[str, str]) -> int:
    return sum(1 for k in collected_pii if k in VERIFICATION_FIELDS)


def pii_value_problem(field_name: str, value: str) -> str | None:
    """Why `value` cannot be a `field_name`, or None if it is plausible.

    The model sometimes records a field from a message that contains no PII
    at all ("I'm calling about my denied claim from January" produced a
    record_pii call), and small models pass the caller's phrasing through
    verbatim. Junk in collected_pii inflates the count that gates
    verify_identity and can shadow a real value given later, so it is
    refused at the door with a reason the model can act on.
    """
    v = (value or "").strip()
    if field_name not in PII_FIELDS:
        return f"unknown field; use one of {', '.join(PII_FIELDS)}"
    if not v:
        return "empty value"
    if field_name == "full_name" and not re.search(r"[A-Za-z]", v):
        return "not a name"
    if field_name == "dob" and _normalize_date(v) is None:
        return "not a recognizable date — re-record it as YYYY-MM-DD"
    if field_name == "phone" and len(re.sub(r"\D", "", v)) < 7:
        return "not a phone number"
    if field_name == "email" and "@" not in v:
        return "not an email address"
    if field_name == "id_last4" and len(re.sub(r"\D", "", v)) != 4:
        return "must be exactly the last 4 digits"
    if field_name == "policy_number" and not re.search(r"\d", v):
        return "not a policy number"
    return None


def match_policyholder(
    collected_pii: dict[str, str],
    policyholders: list[dict],
) -> tuple[dict | None, int]:
    best_match = None
    best_count = 0

    for ph in policyholders:
        count = 0

        if "full_name" in collected_pii:
            names = [ph["name"].lower()]
            names.extend(a.lower() for a in ph.get("name_aliases", []))
            if collected_pii["full_name"].strip().lower() in names:
                count += 1

        if "dob" in collected_pii:
            normalized = _normalize_date(collected_pii["dob"])
            if normalized and normalized == ph.get("dob"):
                count += 1

        if "phone" in collected_pii:
            phones = [_normalize_phone(ph["phone"])]
            phones.extend(_normalize_phone(p) for p in ph.get("phone_aliases", []))
            if _normalize_phone(collected_pii["phone"]) in phones:
                count += 1

        if "email" in collected_pii:
            emails = [ph["email"].lower()]
            emails.extend(e.lower() for e in ph.get("email_aliases", []))
            if collected_pii["email"].strip().lower() in emails:
                count += 1

        if "id_last4" in collected_pii:
            if re.sub(r"\D", "", collected_pii["id_last4"]) == ph.get("id_last4", ""):
                count += 1

        # Deliberately not counted — see VERIFICATION_FIELDS.
        if count > best_count:
            best_count = count
            best_match = ph

    if best_count >= 3:
        return best_match, best_count
    return None, best_count


# ---------------------------------------------------------------------------
# Claim lookups
# ---------------------------------------------------------------------------

def get_claims_for_party(party_id: str, claims: list[dict]) -> list[dict]:
    return [c for c in claims if c["party_id"] == party_id]


def get_claim_by_id(case_id: str, claims: list[dict]) -> dict | None:
    for c in claims:
        if c["case_id"] == case_id:
            return c
    return None


# ---------------------------------------------------------------------------
# Document and follow-up guidance
# ---------------------------------------------------------------------------

def get_document_guidance(document_name: str, guidelines: dict, case_type: str | None = None) -> str:
    doc_guidance = guidelines.get("document_guidance", {})
    alt_guidance = guidelines.get("document_alternative_guidance", {})
    doc_name_lower = document_name.lower().strip()

    matched_key = None
    for key in doc_guidance:
        if doc_name_lower in key.lower() or key.lower() in doc_name_lower:
            matched_key = key
            break

    parts = []
    if matched_key:
        parts.append(f"Submission guidance for '{matched_key}':\n{doc_guidance[matched_key].get('en', '')}")
        if matched_key in alt_guidance:
            parts.append(f"\nIf not available:\n{alt_guidance[matched_key].get('en', '')}")
    else:
        default = guidelines.get("default_guidance", {}).get("en", "")
        parts.append(f"General document submission guidance:\n{default}")
        default_alt = alt_guidance.get("default", {}).get("en", "")
        if default_alt:
            parts.append(f"\nIf not available:\n{default_alt}")

    # Only the note for this claim's own type — emitting every case type gave
    # healthcare callers auto-claim advice.
    case_type_guidance = guidelines.get("case_type_guidance", {})
    for ct, g in case_type_guidance.items():
        if case_type and ct.lower() != case_type.lower():
            continue
        parts.append(f"\n{ct.title()} note: {g.get('en', '')}")

    return "\n".join(parts)


def match_followup_guidance(
    question: str,
    intent: str | None,
    active_claim: dict | None,
    guidelines: dict,
) -> str:
    question_lower = question.lower()
    settings = guidelines.get("claim_followup_settings", {})

    # Preference order:
    #   1. keyword hit on an entry whose intent_hints include this intent
    #   2. keyword hit on any entry — if the caller literally asked about the
    #      topic, that beats the classified intent (a denial_question caller
    #      still deserves submission_method guidance when they ask how to send)
    #   3. an intent-matching catch-all (entries with no `match_any`)
    # Without this, the catch-all listed first shadows every specific rule.
    by_intent_and_keyword = None
    by_keyword = None
    by_intent = None

    for entry in guidelines.get("claim_followup_guidance", []):
        if entry.get("requires_documents") and not (active_claim and active_claim.get("documents_needed")):
            continue

        intent_ok = not intent or intent in entry.get("intent_hints", [])
        match_any = entry.get("match_any", [])

        if match_any and any(kw.lower() in question_lower for kw in match_any):
            if intent_ok:
                by_intent_and_keyword = by_intent_and_keyword or entry
            else:
                by_keyword = by_keyword or entry
        elif not match_any and intent_ok:
            by_intent = by_intent or entry

    entry = by_intent_and_keyword or by_keyword or by_intent
    if entry is None:
        return guidelines.get("claim_followup_fallback", {}).get("en", "No specific guidance available.")

    text = entry["en"]
    if active_claim:
        text = text.replace("{case_id}", active_claim.get("case_id", ""))
        docs = ", ".join(active_claim.get("documents_needed", []))
        text = text.replace("{documents}", docs)
    avg_time = settings.get("average_processing_time_after_submission", {}).get("en", "")
    return text.replace("{average_processing_time_after_submission}", avg_time)