"""Download ten public synthetic records, not real customer records.
Python 3.10+, standard library only. No model call. Not a financial adviser.
Dataset: Elith & NVIDIA, CC BY 4.0. https://creativecommons.org/licenses/by/4.0/
"""
import argparse
import hashlib
import json
from pathlib import Path
from datetime import datetime, timezone
from urllib.parse import urlencode
from urllib.request import urlopen

DATASET = "nvidia/Nemotron-Personas-Finance-Japan"
SOURCE = "https://huggingface.co/datasets/" + DATASET
FIELDS = ("uuid", "age", "prefecture", "financial_literacy", "has_smartphone",
          "finance_persona", "financial_behavior_persona", "financial_concerns_persona")

def validate(payload):
    rows = payload.get("rows")
    if not isinstance(rows, list) or len(rows) != 10:
        raise ValueError("Expected exactly 10 rows; stop and inspect the response")
    records = []
    seen = set()
    for item in rows:
        if item.get("truncated_cells") != []:
            raise ValueError("Missing truncation status or truncated text")
        row = item.get("row")
        if not isinstance(row, dict) or any(k not in row for k in FIELDS):
            raise ValueError("Required fields missing")
        if type(row["age"]) is not int or row["age"] < 18:
            raise ValueError("Expected adult age as an integer")
        if any(not isinstance(row[k], str) or not row[k].strip()
               for k in FIELDS if k != "age"):
            raise ValueError("Empty or invalid text field")
        if row["uuid"] in seen:
            raise ValueError("Duplicate uuid")
        seen.add(row["uuid"])
        records.append({k: row[k] for k in FIELDS})
    return records

def download(url):
    with urlopen(url, timeout=40) as response:
        return response.read()

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--output", default="personas-sample")
    args = parser.parse_args()
    out = Path(args.output)
    if out.exists():
        raise SystemExit("Output already exists; choose a new directory")
    query = urlencode(dict(dataset=DATASET, config="default", split="train", offset=0, length=10))
    url = "https://datasets-server.huggingface.co/rows?" + query
    meta = json.loads(download("https://huggingface.co/api/datasets/" + DATASET))
    raw = download(url)
    payload = json.loads(raw)
    cases = validate(payload)
    manifest = {
        "source": SOURCE, "creators": "Elith & NVIDIA Corporation",
        "license": "CC BY 4.0", "license_url": "https://creativecommons.org/licenses/by/4.0/",
        "retrieved_at_utc": datetime.now(timezone.utc).isoformat(),
        "repository_sha_observed_before_fetch": meta.get("sha"),
        "query_url": url, "response_sha256": hashlib.sha256(raw).hexdigest(),
        "bytes": len(raw), "rows": len(cases), "fields": len(payload.get("features", [])),
        "total_rows_reported": payload.get("num_rows_total"),
        "changes": "First 10 rows, eight fields selected; text unchanged. Synthetic data.",
        "limitations": "Sequential slice, not random. Viewer response is not revision-pinned. No model evaluation."
    }
    out.mkdir(parents=True)
    (out / "rows.json").write_bytes(raw)
    for name, value in [("eval-cases.json", cases), ("manifest.json", manifest)]:
        (out / name).write_text(json.dumps(value, ensure_ascii=False, indent=2), encoding="utf-8")
    print(json.dumps({k: manifest[k] for k in ("rows", "fields", "total_rows_reported", "bytes")}, ensure_ascii=False))

if __name__ == "__main__":
    main()
