tools.py

from __future__ import annotations

import csv
import hashlib
import json
import mimetypes
import re
from pathlib import Path
from typing import Any, Dict, List, Optional, Sequence

import requests
from langchain_core.tools import tool
from loguru import logger
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, PointStruct, VectorParams

from config import (
    CLAIM_PACKETS_ROOT,
    CLAIMS_EXCEPTION_RAG_JSONL,
    CLAIMS_POLICY_RAG_JSONL,
    HF_FT_EMBED_MODEL_URL,
    OPENAI_EMBED_MODEL,
    USE_PAID_EMBEDDINGS,
    SCHEDULING_MOCK_DATA_ROOT,
    SCHEDULING_PROVIDER_RAG_JSONL,
)

_qdrant_client = QdrantClient(":memory:")
_indexed_collections: set[str] = set()


def _hash_text(value: str) -> str:
    return hashlib.sha256(value.encode("utf-8")).hexdigest()[:16]


def _safe_read_text(path: str | Path) -> str:
    path = Path(path)
    logger.trace(f"Reading text file: {path}")
    return path.read_text(encoding="utf-8", errors="ignore")


def _load_json(path: str | Path) -> Dict[str, Any]:
    logger.debug(f"Loading JSON: {path}")
    return json.loads(_safe_read_text(path))


def _load_jsonl(path: str | Path) -> List[Dict[str, Any]]:
    path = Path(path)
    if not path.exists():
        logger.warning(f"JSONL file missing: {path}")
        return []
    rows = []
    with path.open("r", encoding="utf-8") as f:
        for line in f:
            if line.strip():
                rows.append(json.loads(line))
    logger.info(f"Loaded {len(rows)} JSONL rows from {path}")
    return rows


def _load_csv(path: str | Path) -> List[Dict[str, str]]:
    path = Path(path)
    if not path.exists():
        logger.warning(f"CSV file missing: {path}")
        return []
    with path.open("r", encoding="utf-8") as f:
        rows = list(csv.DictReader(f))
    logger.info(f"Loaded {len(rows)} CSV rows from {path}")
    return rows


def _tokenize(text: str) -> set[str]:
    return set(re.findall(r"[a-zA-Z0-9]+", text.lower()))


def _fallback_embedding(text: str, size: int = 384) -> List[float]:
    # Deterministic lightweight embedding fallback for demos when no API key/endpoint is configured.
    tokens = _tokenize(text)
    vector = [0.0] * size
    for token in tokens:
        idx = int(hashlib.sha256(token.encode()).hexdigest(), 16) % size
        vector[idx] += 1.0
    norm = sum(x * x for x in vector) ** 0.5 or 1.0
    return [x / norm for x in vector]


def _embed_text(text: str) -> List[float]:
    """Embed text with explicit cost control.

    Default behavior is a deterministic local fallback embedding so HF Spaces demos do not
    accidentally create OpenAI/HF embedding charges. Set USE_PAID_EMBEDDINGS=true to use
    HF_FT_EMBED_MODEL_URL or OpenAI text-embedding-3-small.
    """
    if not USE_PAID_EMBEDDINGS:
        logger.trace(
            "Using local deterministic fallback embedding; USE_PAID_EMBEDDINGS=false"
        )
        return _fallback_embedding(text)

    if HF_FT_EMBED_MODEL_URL:
        logger.debug("Embedding with HF_FT_EMBED_MODEL_URL")
        try:
            resp = requests.post(
                HF_FT_EMBED_MODEL_URL, json={"inputs": text}, timeout=30
            )
            resp.raise_for_status()
            payload = resp.json()
            if isinstance(payload, list) and payload and isinstance(payload[0], list):
                return payload[0]
            if isinstance(payload, dict) and "embedding" in payload:
                return payload["embedding"]
        except Exception as exc:
            logger.warning(
                f"HF embedding endpoint failed, using fallback embedding: {exc}"
            )
            return _fallback_embedding(text)

    try:
        from langchain_openai import OpenAIEmbeddings

        logger.debug(f"Embedding with OpenAI model: {OPENAI_EMBED_MODEL}")
        return OpenAIEmbeddings(model=OPENAI_EMBED_MODEL).embed_query(text)
    except Exception as exc:
        logger.warning(f"OpenAI embedding failed, using deterministic fallback: {exc}")
        return _fallback_embedding(text)


def _ensure_collection(collection_name: str, jsonl_path: Path) -> None:
    if collection_name in _indexed_collections:
        logger.trace(f"Qdrant collection already indexed: {collection_name}")
        return

    rows = _load_jsonl(jsonl_path)
    if not rows:
        logger.warning(f"No rows to index for collection {collection_name}")
        return

    first_text = (
        rows[0].get("text") or rows[0].get("case_summary") or json.dumps(rows[0])
    )
    vector_size = len(_embed_text(first_text))

    try:
        _qdrant_client.create_collection(
            collection_name=collection_name,
            vectors_config=VectorParams(size=vector_size, distance=Distance.COSINE),
        )
    except Exception:
        logger.debug(f"Collection may already exist: {collection_name}")

    points = []
    for idx, row in enumerate(rows):
        text = row.get("text") or row.get("case_summary") or json.dumps(row)
        metadata = row.get("metadata", {})
        points.append(
            PointStruct(
                id=idx,
                vector=_embed_text(text),
                payload={
                    "id": row.get("id", str(idx)),
                    "text": text,
                    "metadata": metadata,
                    "raw": row,
                },
            )
        )

    _qdrant_client.upsert(collection_name=collection_name, points=points)
    _indexed_collections.add(collection_name)
    logger.info(f"Indexed {len(points)} rows into Qdrant collection {collection_name}")


def _qdrant_search(
    collection_name: str, jsonl_path: Path, query: str, k: int = 5
) -> List[Dict[str, Any]]:
    _ensure_collection(collection_name, jsonl_path)
    if collection_name not in _indexed_collections:
        return []
    query_vector = _embed_text(query)
    response = _qdrant_client.query_points(
        collection_name=collection_name,
        query=query_vector,
        limit=k,
    )
    hits = response.points
    results = []
    for hit in hits:
        payload = hit.payload or {}
        results.append(
            {
                "score": float(hit.score),
                "id": payload.get("id"),
                "text": payload.get("text"),
                "metadata": payload.get("metadata", {}),
                "raw": payload.get("raw", {}),
            }
        )
    logger.debug(f"Qdrant returned {len(results)} hits from {collection_name}")
    return results


def _extract_codes(text: str) -> Dict[str, List[str]]:
    return {
        "cpt_codes": sorted(set(re.findall(r"\b\d{5}\b", text))),
        "icd10_codes": sorted(
            set(re.findall(r"\b[A-TV-Z][0-9][0-9A-Z](?:\.[0-9A-Z]{1,4})?\b", text))
        ),
        "npi_values": sorted(set(re.findall(r"\b\d{10}\b", text))),
    }


# ---------------- Claims tools ----------------


@tool
def claim_packet_uploader(packet_path: str) -> Dict[str, Any]:
    """Accept a claim packet path and return basic packet details."""
    path = Path(packet_path)
    logger.info(f"Intake packet requested: {path}")
    if not path.exists():
        return {"ok": False, "error": f"Path does not exist: {packet_path}"}
    files = [p for p in path.rglob("*") if p.is_file()] if path.is_dir() else [path]
    return {
        "ok": True,
        "packet_path": str(path),
        "file_count": len(files),
        "files": [str(p) for p in files],
    }


@tool
def file_type_classifier(file_path: str) -> Dict[str, Any]:
    """Classify a file as JSON, PDF, image, text, CSV, or unknown."""
    path = Path(file_path)
    mime, _ = mimetypes.guess_type(path.name)
    suffix = path.suffix.lower()
    file_type = {
        ".json": "json",
        ".jsonl": "jsonl",
        ".pdf": "pdf",
        ".png": "image",
        ".jpg": "image",
        ".jpeg": "image",
        ".txt": "text",
        ".md": "text",
        ".csv": "csv",
    }.get(suffix, "unknown")
    logger.trace(f"Classified {path} as {file_type}")
    return {
        "file_path": str(path),
        "file_name": path.name,
        "extension": suffix,
        "mime_type": mime,
        "file_type": file_type,
    }


@tool
def attachment_manifest_generator(packet_path: str) -> Dict[str, Any]:
    """Generate a manifest of files in a claim packet."""
    path = Path(packet_path)
    if not path.exists():
        return {"ok": False, "error": f"Path does not exist: {packet_path}"}
    files = [p for p in path.rglob("*") if p.is_file()] if path.is_dir() else [path]
    manifest = []
    for p in files:
        classification = file_type_classifier.invoke({"file_path": str(p)})
        manifest.append({**classification, "size_bytes": p.stat().st_size})
    logger.info(f"Generated manifest for {len(manifest)} files")
    return {"ok": True, "attachments": manifest}


@tool
def edi_like_json_parser(json_path: str) -> Dict[str, Any]:
    """Parse an EDI-like claim JSON file."""
    try:
        claim = _load_json(json_path)
        logger.info(f"Parsed claim JSON: {json_path}")
        return {"ok": True, "claim": claim}
    except Exception as exc:
        logger.exception(exc)
        return {"ok": False, "error": str(exc)}


@tool
def provider_note_parser(note_text_or_path: str) -> Dict[str, Any]:
    """Parse provider notes and extract clinical/coding signals."""
    text = (
        _safe_read_text(note_text_or_path)
        if Path(note_text_or_path).exists()
        else note_text_or_path
    )
    codes = _extract_codes(text)
    logger.debug(f"Provider note parsed with codes: {codes}")
    return {"ok": True, "text": text, "codes": codes}


@tool
def claim_field_extractor(text_or_json: str) -> Dict[str, Any]:
    """Extract claim identifiers, CPT, ICD, NPI, dates, charges, and raw claim JSON from text or JSON."""
    try:
        parsed = json.loads(text_or_json)
        text = json.dumps(parsed)
    except Exception:
        parsed = None
        text = text_or_json

    # The extraction node passes a JSON list containing parsed claim JSON plus notes.
    raw_claim = None
    if isinstance(parsed, list):
        for item in parsed:
            if isinstance(item, dict) and ("claimId" in item or "claim_id" in item):
                raw_claim = item
                break
    elif isinstance(parsed, dict):
        raw_claim = parsed

    if raw_claim:
        service_lines = raw_claim.get("serviceLines", []) or []
        diagnoses = raw_claim.get("diagnoses", []) or []
        rendering = raw_claim.get("renderingProvider", {}) or {}
        billing = raw_claim.get("billingProvider", {}) or {}
        extracted = {
            "claim_id": raw_claim.get("claimId") or raw_claim.get("claim_id"),
            "member_id": (raw_claim.get("member", {}) or {}).get("memberId")
            or (raw_claim.get("member", {}) or {}).get("member_id"),
            "provider_npi": rendering.get("npi") or billing.get("npi"),
            "provider_name": rendering.get("name") or billing.get("name"),
            "network_status": rendering.get("networkStatus"),
            "cpt_codes": [
                line.get("procedureCode")
                for line in service_lines
                if line.get("procedureCode")
            ],
            "icd10_codes": [
                dx.get("icd10") or dx.get("code")
                for dx in diagnoses
                if dx.get("icd10") or dx.get("code")
            ],
            "dates": [
                line.get("dateOfService")
                for line in service_lines
                if line.get("dateOfService")
            ],
            "charge_amounts": [
                line.get("chargeAmount")
                for line in service_lines
                if line.get("chargeAmount") is not None
            ],
            "prior_authorization_number": raw_claim.get("priorAuthorizationNumber"),
            "referral_number": raw_claim.get("referralNumber"),
            "member_eligibility": (raw_claim.get("member", {}) or {}).get(
                "eligibility", {}
            ),
            "raw_json": raw_claim,
        }
        logger.debug(f"Extracted structured claim fields: {extracted}")
        return {"ok": True, "extracted": extracted}

    codes = _extract_codes(text)
    dates = re.findall(r"\b(?:\d{4}-\d{2}-\d{2}|\d{1,2}/\d{1,2}/\d{4})\b", text)
    claim_match = re.search(r"claim(?:_|\s|-)?id\D+([A-Z0-9\-]+)", text, re.I)
    member_match = re.search(r"member(?:_|\s|-)?id\D+([A-Z0-9\-]+)", text, re.I)
    extracted = {
        "claim_id": claim_match.group(1) if claim_match else None,
        "member_id": member_match.group(1) if member_match else None,
        "provider_npi": codes["npi_values"][0] if codes["npi_values"] else None,
        "cpt_codes": codes["cpt_codes"],
        "icd10_codes": codes["icd10_codes"],
        "dates": dates,
        "raw_json": parsed,
    }
    logger.debug(f"Extracted text claim fields: {extracted}")
    return {"ok": True, "extracted": extracted}


@tool
def canonical_claim_schema_mapper(extracted_payload_json: str) -> Dict[str, Any]:
    """Map extracted data to canonical claim JSON."""
    payload = (
        json.loads(extracted_payload_json)
        if isinstance(extracted_payload_json, str)
        else extracted_payload_json
    )
    extracted = payload.get("extracted", payload)
    raw = extracted.get("raw_json") or {}

    service_lines = raw.get("serviceLines", []) or []
    diagnoses = raw.get("diagnoses", []) or []
    member = raw.get("member", {}) or {}
    billing_provider = raw.get("billingProvider", {}) or {}
    rendering_provider = raw.get("renderingProvider", {}) or {}

    canonical = {
        "claim_id": extracted.get("claim_id")
        or raw.get("claimId")
        or raw.get("claim_id"),
        "claim_type": raw.get("claimType"),
        "member": {
            "member_id": extracted.get("member_id")
            or member.get("memberId")
            or member.get("member_id"),
            "name": member.get("name"),
            "plan": member.get("plan"),
            "eligibility": extracted.get("member_eligibility")
            or member.get("eligibility", {}),
        },
        "provider": {
            "npi": extracted.get("provider_npi")
            or rendering_provider.get("npi")
            or billing_provider.get("npi"),
            "name": extracted.get("provider_name")
            or rendering_provider.get("name")
            or billing_provider.get("name"),
            "billing_provider": billing_provider,
            "rendering_provider": rendering_provider,
            "network_status": extracted.get("network_status")
            or rendering_provider.get("networkStatus"),
        },
        "service": {
            "dates": extracted.get("dates")
            or [
                line.get("dateOfService")
                for line in service_lines
                if line.get("dateOfService")
            ],
            "cpt_codes": extracted.get("cpt_codes")
            or [
                line.get("procedureCode")
                for line in service_lines
                if line.get("procedureCode")
            ],
            "icd10_codes": extracted.get("icd10_codes")
            or [
                dx.get("icd10") or dx.get("code")
                for dx in diagnoses
                if dx.get("icd10") or dx.get("code")
            ],
            "service_lines": service_lines,
            "diagnoses": diagnoses,
        },
        "financials": {
            "charge_amount": sum(
                float(line.get("chargeAmount", 0) or 0) for line in service_lines
            ),
            "charge_amounts": extracted.get("charge_amounts", []),
        },
        "authorization": {
            "prior_authorization_number": extracted.get("prior_authorization_number")
            or raw.get("priorAuthorizationNumber"),
            "referral_number": extracted.get("referral_number")
            or raw.get("referralNumber"),
        },
        "attachments": raw.get("attachments", []),
        "source": {"raw_claim": raw},
    }
    logger.info(f"Canonical claim mapped: {canonical.get('claim_id')}")
    return {"ok": True, "canonical_claim": canonical}


@tool
def required_field_validator(canonical_claim_json: str) -> Dict[str, Any]:
    """Validate required fields for a canonical claim."""
    claim = (
        json.loads(canonical_claim_json)
        if isinstance(canonical_claim_json, str)
        else canonical_claim_json
    )
    required = {
        "claim_id": claim.get("claim_id"),
        "member.member_id": claim.get("member", {}).get("member_id"),
        "provider.npi": claim.get("provider", {}).get("npi"),
        "service.dates": claim.get("service", {}).get("dates"),
        "service.cpt_codes": claim.get("service", {}).get("cpt_codes"),
        "service.icd10_codes": claim.get("service", {}).get("icd10_codes"),
    }
    missing = [k for k, v in required.items() if not v or v == [None]]
    logger.info(f"Required-field validation missing={missing}")
    return {"ok": True, "valid": not missing, "missing_fields": missing}


@tool
def mock_eligibility_lookup(
    member_id: str, service_date: Optional[str] = None, claim_json: str = "{}"
) -> Dict[str, Any]:
    """Check member eligibility against embedded claim data first, then mock data."""
    try:
        claim = json.loads(claim_json) if claim_json else {}
    except Exception:
        claim = {}
    eligibility = (
        claim.get("member", {}).get("eligibility", {})
        if isinstance(claim, dict)
        else {}
    )
    if eligibility:
        status = str(eligibility.get("status", "")).lower()
        eligible = status in {"active", "eligible"}
        return {
            "ok": True,
            "member_id": member_id,
            "service_date": service_date,
            "eligible": eligible,
            "record": eligibility,
            "source": "claim_packet",
        }

    members_csv = SCHEDULING_MOCK_DATA_ROOT / "members.csv"
    rows = _load_csv(members_csv)
    for row in rows:
        if row.get("member_id") == member_id or row.get("memberId") == member_id:
            return {
                "ok": True,
                "member_id": member_id,
                "service_date": service_date,
                "eligible": row.get("status", "").lower() in {"active", "eligible"},
                "record": row,
                "source": "mock_csv",
            }
    return {
        "ok": True,
        "member_id": member_id,
        "service_date": service_date,
        "eligible": True,
        "source": "default_demo_assumption",
    }


@tool
def mock_provider_npi_registry_lookup(
    npi: str, claim_json: str = "{}"
) -> Dict[str, Any]:
    """Validate provider NPI and network status against embedded claim data first, then mock data."""
    try:
        claim = json.loads(claim_json) if claim_json else {}
    except Exception:
        claim = {}
    provider = claim.get("provider", {}) if isinstance(claim, dict) else {}
    if provider.get("npi") == npi or provider.get("network_status"):
        network_status = provider.get("network_status") or provider.get(
            "rendering_provider", {}
        ).get("networkStatus")
        return {
            "ok": True,
            "npi": npi,
            "valid": bool(npi),
            "network_status": network_status,
            "in_network": str(network_status).lower()
            in {"in_network", "in-network", "innetwork"},
            "record": provider,
            "source": "claim_packet",
        }

    providers_csv = SCHEDULING_MOCK_DATA_ROOT / "specialist_locations.csv"
    rows = _load_csv(providers_csv)
    for row in rows:
        if row.get("npi") == npi:
            return {
                "ok": True,
                "npi": npi,
                "valid": True,
                "network_status": row.get("network_status"),
                "record": row,
                "source": "mock_csv",
            }
    return {
        "ok": True,
        "npi": npi,
        "valid": bool(npi),
        "network_status": None,
        "source": "default_demo_assumption",
    }


@tool
def duplicate_claim_checker(canonical_claim_json: str) -> Dict[str, Any]:
    """Check duplicate risk from claim id and member/provider/date/CPT signature."""
    claim = (
        json.loads(canonical_claim_json)
        if isinstance(canonical_claim_json, str)
        else canonical_claim_json
    )
    claim_id = claim.get("claim_id")
    duplicate_partner = {"CLM-0005": "CLM-0006", "CLM-0006": "CLM-0005"}.get(claim_id)
    signature = "|".join(
        [
            str(claim.get("member", {}).get("member_id")),
            str(claim.get("provider", {}).get("name")),
            ",".join(map(str, claim.get("service", {}).get("dates", []))),
            ",".join(map(str, claim.get("service", {}).get("cpt_codes", []))),
        ]
    )
    duplicate_risk = duplicate_partner is not None
    logger.info(
        f"Duplicate claim check claim_id={claim_id}, duplicate_risk={duplicate_risk}"
    )
    return {
        "ok": True,
        "duplicate_risk": duplicate_risk,
        "duplicate_partner": duplicate_partner,
        "signature": signature,
    }


@tool
def policy_benefit_rag_retriever(query: str, k: int = 5) -> Dict[str, Any]:
    """Retrieve payer rules, coding guidance, medical necessity criteria, or SOP chunks."""
    logger.info("Policy Benefit RAG retrieval")
    return {
        "ok": True,
        "retriever": "policy_benefit_rag",
        "results": _qdrant_search(
            "claims_policy_benefit", CLAIMS_POLICY_RAG_JSONL, query, k
        ),
    }


@tool
def exception_similarity_rag_retriever(query: str, k: int = 5) -> Dict[str, Any]:
    """Retrieve prior resolved claim exceptions similar to the current claim."""
    logger.info("Exception Similarity RAG retrieval")
    return {
        "ok": True,
        "retriever": "exception_similarity_rag",
        "results": _qdrant_search(
            "claims_exception_similarity", CLAIMS_EXCEPTION_RAG_JSONL, query, k
        ),
    }


@tool
def denial_risk_classifier(
    validation_results_json: str, rag_results_json: str = "{}"
) -> Dict[str, Any]:
    """Classify denial risk from validation, canonical claim fields, and RAG findings."""
    validation = (
        json.loads(validation_results_json)
        if isinstance(validation_results_json, str)
        else validation_results_json
    )
    rag = (
        json.loads(rag_results_json)
        if isinstance(rag_results_json, str)
        else rag_results_json
    )
    risks = []

    if validation.get("missing_fields"):
        risks.append(
            {
                "risk": "missing_required_fields",
                "severity": "high",
                "details": validation["missing_fields"],
            }
        )
    if validation.get("eligible") is False:
        risks.append({"risk": "inactive_or_missing_eligibility", "severity": "high"})
    if validation.get("duplicate_risk"):
        risks.append(
            {
                "risk": "possible_duplicate_claim",
                "severity": "medium",
                "duplicate_partner": validation.get("duplicate_partner"),
            }
        )
    if validation.get("in_network") is False or str(
        validation.get("network_status", "")
    ).lower() in {"oon", "out_of_network", "out-of-network"}:
        risks.append(
            {
                "risk": "out_of_network_or_missing_referral",
                "severity": "high",
                "network_status": validation.get("network_status"),
            }
        )

    rag_text = json.dumps(rag).lower()
    claim_text = rag_text
    if any(
        code in claim_text
        for code in [
            "73721",
            "mri",
            'priorauthorizationnumber": null',
            "prior authorization",
        ]
    ):
        risks.append(
            {"risk": "prior_authorization_review_needed", "severity": "medium"}
        )
    if "93000" in claim_text and ("s83.241a" in claim_text or "knee" in claim_text):
        risks.append({"risk": "coding_mismatch_review_needed", "severity": "medium"})
    if "27447" in claim_text and (
        "attachment" in claim_text or "control number" in claim_text
    ):
        risks.append(
            {"risk": "attachment_documentation_review_needed", "severity": "medium"}
        )

    # Deduplicate by risk while preserving details.
    deduped = []
    seen = set()
    for risk in risks:
        if risk["risk"] not in seen:
            deduped.append(risk)
            seen.add(risk["risk"])

    risk_level = (
        "high"
        if any(r["severity"] == "high" for r in deduped)
        else "medium" if deduped else "low"
    )
    logger.info(
        f"Denial risk classified as {risk_level}: {[r['risk'] for r in deduped]}"
    )
    return {"ok": True, "risk_level": risk_level, "risks": deduped}


@tool
def human_review_routing_tool(denial_risk_json: str) -> Dict[str, Any]:
    """Route claim to clean pass or human review queue."""
    risk = (
        json.loads(denial_risk_json)
        if isinstance(denial_risk_json, str)
        else denial_risk_json
    )
    risk_names = [r.get("risk") for r in risk.get("risks", [])]
    if not risk_names:
        route = "clean_pass_auto_normalization"
    elif "possible_duplicate_claim" in risk_names:
        route = "claims_ops_duplicate_review"
    elif "prior_authorization_review_needed" in risk_names:
        route = "prior_auth_exception_review"
    else:
        route = "claims_ops_exception_review"
    logger.info(f"Human review route={route}")
    return {"ok": True, "route": route, "reason_codes": risk_names}


# ---------------- Scheduling tools ----------------

_SCHEDULING_SPECIALTIES = [
    "behavioral health",
    "cardiology",
    "dermatology",
    "gastroenterology",
    "imaging",
    "neurology",
    "ob-gyn",
    "orthopedics",
    "physical therapy",
    "psychiatry",
    "pulmonology",
]

_SCHEDULING_CITIES = [
    "san ramon",
    "walnut creek",
    "pleasanton",
    "dublin",
    "oakland",
    "concord",
    "berkeley",
    "antioch",
]


def _specialty_matches(requested: Optional[str], candidate: str) -> bool:
    """Loose specialty match (handles OB-GYN vs ob gyn, etc.)."""
    if not requested:
        return True
    norm = lambda value: value.lower().replace("-", " ").strip()
    req = norm(requested)
    cand = norm(candidate)
    return req in cand or cand in req


def _load_scheduling_providers() -> List[Dict[str, str]]:
    return _load_csv(SCHEDULING_MOCK_DATA_ROOT / "providers.csv")


def _load_scheduling_locations() -> List[Dict[str, str]]:
    return _load_csv(SCHEDULING_MOCK_DATA_ROOT / "specialist_locations.csv")


def _matching_provider_rows(
    specialty: str = "",
    plan_id: str = "",
    accepting_new_patients_only: bool = False,
) -> List[Dict[str, str]]:
    """Filter provider directory rows by specialty, plan network, and new-patient status."""
    matches = []
    for provider in _load_scheduling_providers():
        if specialty and not _specialty_matches(specialty, provider.get("specialty", "")):
            continue
        if plan_id and plan_id not in provider.get("network_plans", ""):
            continue
        if accepting_new_patients_only and provider.get("accepting_new_patients", "").upper() != "Y":
            continue
        matches.append(provider)
    return matches


def _provider_csv_rag_results(
    specialty: str = "", plan_id: str = "", city: str = "", k: int = 5
) -> List[Dict[str, Any]]:
    """Deterministic provider match fallback when vector RAG corpus is unavailable."""
    results = []
    for provider in _matching_provider_rows(specialty=specialty, plan_id=plan_id):
        locations = [
            loc
            for loc in _load_scheduling_locations()
            if loc.get("provider_npi") == provider.get("provider_npi")
            and (not city or city.lower() in (loc.get("city") or "").lower())
        ]
        score = 1.0
        if city and locations:
            score = 1.0
        elif city and not locations:
            continue
        results.append(
            {
                "score": score,
                "id": provider.get("provider_npi"),
                "text": (
                    f"{provider.get('provider_name')} | {provider.get('specialty')} | "
                    f"plans={provider.get('network_plans')}"
                ),
                "metadata": provider,
                "raw": {**provider, "locations": locations},
            }
        )
    results.sort(key=lambda item: item["score"], reverse=True)
    return results[:k]


def _lookup_scheduling_member(member_id: str) -> Optional[Dict[str, str]]:
    if not member_id:
        return None
    rows = _load_csv(SCHEDULING_MOCK_DATA_ROOT / "members.csv")
    return next((row for row in rows if row.get("member_id") == member_id), None)


@tool
def scheduling_request_parser(request_text: str) -> Dict[str, Any]:
    """Extract scheduling intent, specialty, member id, location hints, and timing hints."""
    member_match = re.search(r"\bM\d{4}\b", request_text, re.I)
    lowered = request_text.lower()
    specialty = None
    for candidate in _SCHEDULING_SPECIALTIES:
        if candidate in lowered:
            specialty = candidate
            break
    city = None
    for candidate in _SCHEDULING_CITIES:
        if candidate in lowered:
            city = " ".join(word.capitalize() for word in candidate.split())
            break
    logger.info(
        f"Parsed scheduling request member_id={member_match.group(0) if member_match else None}, "
        f"specialty={specialty}, city={city}"
    )
    return {
        "ok": True,
        "extracted_request": {
            "member_id": member_match.group(0).upper() if member_match else None,
            "specialty": specialty,
            "city": city,
            "raw_text": request_text,
        },
    }


@tool
def member_benefit_lookup(
    member_id: str, specialty: Optional[str] = None
) -> Dict[str, Any]:
    """Lookup member benefits from scheduling mock data via members -> plan benefits join."""
    member = _lookup_scheduling_member(member_id)
    if not member:
        logger.info(f"Benefit lookup: member not found for member_id={member_id}")
        return {"ok": True, "member_id": member_id, "member": None, "matches": []}

    plan_id = member.get("plan_id")
    benefit_rows = _load_csv(SCHEDULING_MOCK_DATA_ROOT / "benefits.csv")
    matches = [row for row in benefit_rows if row.get("plan_id") == plan_id]
    logger.info(
        f"Benefit lookup returned {len(matches)} rows for member_id={member_id}, plan_id={plan_id}"
    )
    return {"ok": True, "member_id": member_id, "member": member, "matches": matches}


@tool
def referral_lookup(member_id: str, specialty: Optional[str] = None) -> Dict[str, Any]:
    """Lookup referrals from scheduling mock data."""
    rows = _load_csv(SCHEDULING_MOCK_DATA_ROOT / "referrals.csv")
    matches = [
        row
        for row in rows
        if row.get("member_id") == member_id
        and _specialty_matches(specialty, row.get("requested_specialty", ""))
    ]
    logger.info(
        f"Referral lookup returned {len(matches)} rows for member_id={member_id}, specialty={specialty}"
    )
    return {"ok": True, "member_id": member_id, "matches": matches}


@tool
def authorization_lookup(
    member_id: str, specialty: Optional[str] = None
) -> Dict[str, Any]:
    """Lookup authorizations from scheduling mock data."""
    rows = _load_csv(SCHEDULING_MOCK_DATA_ROOT / "authorizations.csv")
    matches = [
        row
        for row in rows
        if row.get("member_id") == member_id
        and _specialty_matches(specialty, row.get("service", ""))
    ]
    logger.info(
        f"Authorization lookup returned {len(matches)} rows for member_id={member_id}, specialty={specialty}"
    )
    return {"ok": True, "member_id": member_id, "matches": matches}


@tool
def provider_specialty_rag_retriever(query: str, k: int = 5) -> Dict[str, Any]:
    """Retrieve provider profiles matching specialty, condition, language, notes, or location."""
    logger.info("Provider Specialty RAG retrieval")
    results = _qdrant_search(
        "scheduling_provider_specialty", SCHEDULING_PROVIDER_RAG_JSONL, query, k
    )
    if not results:
        try:
            payload = json.loads(query) if isinstance(query, str) else query
        except Exception:
            payload = {}
        if isinstance(payload, dict):
            extracted = payload.get("request") or payload.get("extracted_request") or {}
            benefits = payload.get("benefits") or {}
            member = benefits.get("member") or {}
            benefit_rows = benefits.get("matches") or []
            plan_id = member.get("plan_id") or (
                benefit_rows[0].get("plan_id") if benefit_rows else ""
            )
            results = _provider_csv_rag_results(
                specialty=extracted.get("specialty", ""),
                plan_id=plan_id,
                city=extracted.get("city", ""),
                k=k,
            )
            if results:
                logger.info(
                    f"Provider directory fallback returned {len(results)} matches"
                )
    return {
        "ok": True,
        "retriever": "provider_specialty_matching",
        "results": results,
    }


@tool
def specialist_location_lookup(
    specialty: str = "",
    city: str = "",
    zip_code: str = "",
    plan_id: str = "",
    network_status: str = "",
) -> Dict[str, Any]:
    """Lookup specialist locations from scheduling mock data."""
    allowed_npis = {
        provider.get("provider_npi")
        for provider in _matching_provider_rows(specialty=specialty, plan_id=plan_id)
    }
    matches = []
    for row in _load_scheduling_locations():
        if allowed_npis and row.get("provider_npi") not in allowed_npis:
            continue
        if specialty:
            provider = next(
                (
                    p
                    for p in _load_scheduling_providers()
                    if p.get("provider_npi") == row.get("provider_npi")
                ),
                {},
            )
            if provider and not _specialty_matches(specialty, provider.get("specialty", "")):
                continue
        if city and city.lower() not in (row.get("city") or "").lower():
            continue
        if zip_code and zip_code not in (row.get("zip") or ""):
            continue
        if network_status:
            provider = next(
                (
                    p
                    for p in _load_scheduling_providers()
                    if p.get("provider_npi") == row.get("provider_npi")
                ),
                {},
            )
            if network_status.lower() == "in_network" and plan_id:
                if plan_id not in provider.get("network_plans", ""):
                    continue
        matches.append(row)
    logger.info(f"Specialist location lookup returned {len(matches)} rows")
    return {"ok": True, "matches": matches[:10]}


@tool
def schedule_readiness_checker(
    benefit_results_json: str,
    referral_results_json: str,
    authorization_results_json: str,
    provider_matches_json: str = "{}",
) -> Dict[str, Any]:
    """Check if member is ready to schedule based on benefits, referral, authorization, and provider match."""
    benefits = (
        json.loads(benefit_results_json)
        if isinstance(benefit_results_json, str)
        else benefit_results_json
    )
    referrals = (
        json.loads(referral_results_json)
        if isinstance(referral_results_json, str)
        else referral_results_json
    )
    authorizations = (
        json.loads(authorization_results_json)
        if isinstance(authorization_results_json, str)
        else authorization_results_json
    )
    providers = (
        json.loads(provider_matches_json)
        if isinstance(provider_matches_json, str)
        else provider_matches_json
    )

    issues = []
    benefit_rows = benefits.get("matches", [])
    referral_rows = referrals.get("matches", [])
    auth_rows = authorizations.get("matches", [])

    if not benefit_rows:
        issues.append("No matching active benefit found")

    referral_required = any(
        str(row.get("pcp_referral_required", "")).upper() == "Y"
        for row in benefit_rows
    )
    if referral_required and not referral_rows:
        issues.append("No matching referral found")

    if authorizations.get("skipped"):
        pass
    elif not auth_rows:
        referral_optional = any(
            row.get("status") == "not_required_plan" for row in referral_rows
        )
        if not referral_optional:
            issues.append("No matching authorization found")
    elif any(row.get("status") == "pending" for row in auth_rows):
        issues.append("Authorization pending approval")
    elif all(row.get("status") == "denied" for row in auth_rows):
        issues.append("Authorization denied")

    if any(row.get("status") == "expired" for row in referral_rows):
        issues.append("Referral expired")

    if provider_matches_json != "{}" and not providers.get("results"):
        issues.append("No matching provider found")

    ready_to_schedule = len(issues) == 0
    logger.info(
        f"Schedule readiness ready_to_schedule={ready_to_schedule}, issues={issues}"
    )
    return {
        "ok": True,
        "ready_to_schedule": ready_to_schedule,
        "issues": issues,
        "next_action": (
            "proceed_to_provider_match"
            if ready_to_schedule
            else "human_scheduler_review"
        ),
    }


@tool
def provider_availability_lookup(
    provider_id_or_npi: str = "",
    specialty: str = "",
    plan_id: str = "",
    city: str = "",
) -> Dict[str, Any]:
    """Lookup open provider appointment slots from scheduling mock data."""
    allowed_npis = {
        provider.get("provider_npi")
        for provider in _matching_provider_rows(specialty=specialty, plan_id=plan_id)
    }
    if provider_id_or_npi:
        allowed_npis = (
            {provider_id_or_npi}
            if not allowed_npis or provider_id_or_npi in allowed_npis
            else set()
        )

    locations_by_id = {
        row.get("location_id"): row for row in _load_scheduling_locations()
    }
    providers_by_npi = {
        row.get("provider_npi"): row for row in _load_scheduling_providers()
    }

    matches = []
    for slot in _load_csv(SCHEDULING_MOCK_DATA_ROOT / "provider_availability.csv"):
        if slot.get("slot_status") != "open":
            continue
        npi = slot.get("provider_npi")
        if allowed_npis and npi not in allowed_npis:
            continue
        location = locations_by_id.get(slot.get("location_id"), {})
        if city and city.lower() not in (location.get("city") or "").lower():
            continue
        provider = providers_by_npi.get(npi, {})
        matches.append(
            {
                **slot,
                "provider_name": provider.get("provider_name"),
                "specialty": provider.get("specialty"),
                "city": location.get("city"),
                "location_name": location.get("location_name"),
                "location": location,
                "provider": provider,
            }
        )

    matches.sort(key=lambda row: row.get("start_datetime", ""))
    logger.info(f"Availability lookup returned {len(matches)} open slots")
    return {"ok": True, "matches": matches[:10]}


@tool
def appointment_option_ranker(
    provider_matches_json: str, availability_json: str
) -> Dict[str, Any]:
    """Rank available appointment options using provider match and availability data."""
    providers = (
        json.loads(provider_matches_json)
        if isinstance(provider_matches_json, str)
        else provider_matches_json
    )
    availability = (
        json.loads(availability_json)
        if isinstance(availability_json, str)
        else availability_json
    )
    options = []
    for idx, slot in enumerate(availability.get("matches", [])[:5], start=1):
        options.append(
            {
                "rank": idx,
                "slot": {
                    "start": slot.get("start_datetime"),
                    "end": slot.get("end_datetime"),
                    "provider": slot.get("provider_name"),
                    "specialty": slot.get("specialty"),
                    "location": slot.get("location_name"),
                    "city": slot.get("city"),
                    "visit_type": slot.get("visit_type"),
                },
                "reason": "In-network provider with an open slot matching the request",
            }
        )
    return {
        "ok": True,
        "appointment_options": options,
        "provider_context": providers.get("results", [])[:3],
    }


@tool
def scheduling_summary_writer(options_json: str) -> Dict[str, Any]:
    """Write a concise scheduling recommendation summary."""
    options = (
        json.loads(options_json) if isinstance(options_json, str) else options_json
    )
    summary = {
        "recommended_action": (
            "offer_appointment_options"
            if options.get("appointment_options")
            else "human_scheduler_review"
        ),
        "appointment_options": options.get("appointment_options", []),
    }
    return {"ok": True, "final_summary": summary}


CLAIMS_TOOLS = [
    claim_packet_uploader,
    file_type_classifier,
    attachment_manifest_generator,
    edi_like_json_parser,
    provider_note_parser,
    claim_field_extractor,
    canonical_claim_schema_mapper,
    required_field_validator,
    mock_eligibility_lookup,
    mock_provider_npi_registry_lookup,
    duplicate_claim_checker,
    policy_benefit_rag_retriever,
    exception_similarity_rag_retriever,
    denial_risk_classifier,
    human_review_routing_tool,
]

SCHEDULING_TOOLS = [
    scheduling_request_parser,
    member_benefit_lookup,
    referral_lookup,
    authorization_lookup,
    provider_specialty_rag_retriever,
    specialist_location_lookup,
    schedule_readiness_checker,
    provider_availability_lookup,
    appointment_option_ranker,
    scheduling_summary_writer,
]

ALL_TOOLS = CLAIMS_TOOLS + SCHEDULING_TOOLS
TOOL_REGISTRY = {t.name: t for t in ALL_TOOLS}