from __future__ import annotations

import json
import os
from pathlib import Path

import numpy as np
import pandas as pd
import pytest


pytestmark = pytest.mark.integration


def test_tensor_bridge_smoke_artifacts_contract() -> None:
    out = os.environ.get("SCTM_TENSOR_BRIDGE_OUTPUT")
    if not out:
        return
    root = Path(out)
    tensor_dir = root / "tensor"
    stage0_dir = root / "stage0"
    manifest = json.loads((root / "tensor_builder_manifest.json").read_text(encoding="utf-8"))
    split_counts = pd.read_csv(root / "tensor_split_counts.csv")
    meta = json.loads((tensor_dir / "tensor_metadata.json").read_text(encoding="utf-8"))
    vocab = json.loads((tensor_dir / "cat_value_vocab.json").read_text(encoding="utf-8"))
    assert "变动情况" not in "".join(meta["cat_cols"])
    assert "失访原因" not in "".join(meta["cat_cols"])
    assert not any("变动情况=" in key or "失访原因=" in key for key in vocab)
    for field in [
        "治疗方式",
        "服药方式",
        "服药时间",
        "是否免费服药",
        "是否转诊",
        "转诊类型",
        "随访类型",
        "本次访视方式",
        "本次访视对象",
        "是否应急处置",
        "健康体检",
        "专科医生意见-治疗",
        "基础管理分级",
    ]:
        assert field in meta["cat_cols"]
    assert "公安滋事总数" in meta["num_cols"]
    assert "公安肇事总数" in meta["num_cols"]
    assert "has_static" in meta["static_cols"]
    assert manifest["n_windowed_patients"] == manifest["n_patients_used"]
    assert {"n_singleton_patients", "n_transition_eligible_windows", "n_windowed_patients"}.issubset(split_counts.columns)
    assert int(split_counts["n_windowed_patients"].sum()) == manifest["n_patients_used"]
    assert int(split_counts["n_transition_eligible_windows"].sum()) > 0
    assert (stage0_dir / "ordinal_direction_table.json").exists()
    assert (stage0_dir / "service_state_transitions_train.json").exists()
    with np.load(Path(meta["split_paths"]["val"]), allow_pickle=True) as z:
        assert z["cat_value_ids"].shape[1:] == (meta["max_seq_len"], len(meta["cat_cols"]))
        assert z["numeric_values"].shape[1:] == (meta["max_seq_len"], len(meta["num_cols"]))
        assert z["ordinal_cbe"].shape[1:] == (meta["max_seq_len"], len(meta["ord_cols"]), meta["cbe_dim"])
        assert z["static_value_ids"].shape[1] == len(meta["static_cols"])
        assert z["valid_mask"].sum() > 0