"""Regression: strong_negative_miss == cell[1][5] and is a strict subset of any_negative.

ChatGPT review 0a6e24af — v1 incorrectly set strong==any via (pago-pexp)>=2 on {1,3,5}.
Strong negative = Better(1)→Worse(5) ONLY.

No M1 scoring / L1 changes. Research-only.
"""
from __future__ import annotations

import json
from pathlib import Path

import pytest

ROOT = Path(__file__).resolve().parents[2]
V2_JSON = ROOT / "data/m1/v2/m1long_v2_selection_ipw_sensitivity.json"
V2_NAMED = ROOT / "data/m1/v2/m1long_v2_selection_ipw_sensitivity-v2.json"
V1_BUGGY = ROOT / "data/m1/v2/m1long_v2_selection_ipw_sensitivity-v1-buggy.json"

EXPECTED_STRONG = {
    "unweighted": 105 / 1424,  # 0.07373595505617977
    # weighted values verified from published 3x3 mass / total
    "ipw_stabilized": 0.08648604054331696,
    "ipw_trunc_1_99": 0.08640382332643122,
    "ipw_trunc_5_95": 0.08579335721272811,
}
EXPECTED_ANY_UNWEIGHTED = 382 / 1424  # 0.26825842696629215


def _load_pack() -> dict:
    for p in (V2_NAMED, V2_JSON):
        if p.exists():
            return json.loads(p.read_text())
    pytest.fail(f"missing corrected pack at {V2_JSON} or {V2_NAMED}")


def _cell15(block: dict) -> float:
    detail = block["detail"]
    shares = detail["shares"]
    return float(shares["1"]["5"])


@pytest.mark.parametrize("spec", list(EXPECTED_STRONG))
def test_strong_equals_cell_1_5(spec: str):
    pack = _load_pack()
    block = pack["three_by_three_pexp_pago"][spec]
    strong = block["share_strong_negative_miss"]
    any_neg = block["share_any_negative_miss"]
    cell = _cell15(block)
    assert strong == pytest.approx(cell, abs=1e-12)
    assert strong == pytest.approx(EXPECTED_STRONG[spec], abs=1e-9)
    assert strong < any_neg
    # subset: strong mass cannot exceed any-negative
    assert strong <= any_neg + 1e-15


def test_any_negative_unweighted_unchanged():
    pack = _load_pack()
    uw = pack["three_by_three_pexp_pago"]["unweighted"]
    assert uw["share_any_negative_miss"] == pytest.approx(EXPECTED_ANY_UNWEIGHTED, abs=1e-12)
    # 1→3 + 1→5 + 3→5 = 92+105+185 = 382
    c = uw["detail"]["counts"]
    assert c["1"]["3"] + c["1"]["5"] + c["3"]["5"] == 382
    assert c["1"]["5"] == 105


def test_v1_buggy_preserved_and_was_wrong():
    assert V1_BUGGY.exists(), "v1 buggy artifact must be preserved"
    buggy = json.loads(V1_BUGGY.read_text())
    for spec, block in buggy["three_by_three_pexp_pago"].items():
        # historical bug identity
        assert block["share_strong_negative_miss"] == pytest.approx(
            block["share_any_negative_miss"], abs=1e-15
        ), spec


def test_summarize_helper_unit():
    """Unit-level guard independent of full microdata rebuild."""
    shares = {
        "1": {"1": 0.24, "3": 0.06, "5": 0.07},
        "3": {"1": 0.20, "3": 0.17, "5": 0.13},
        "5": {"1": 0.02, "3": 0.04, "5": 0.07},
    }
    any_neg = shares["1"]["3"] + shares["1"]["5"] + shares["3"]["5"]
    strong = shares["1"]["5"]
    # emulate fixed rule
    strong_calc = 0.0
    any_calc = 0.0
    for r, cols in shares.items():
        for c, s in cols.items():
            ri, ci = int(r), int(c)
            if ci > ri:
                any_calc += s
            if ri == 1 and ci == 5:
                strong_calc += s
    assert strong_calc == pytest.approx(strong)
    assert any_calc == pytest.approx(any_neg)
    assert strong_calc < any_calc
    # buggy rule would equal any
    buggy = sum(s for r, cols in shares.items() for c, s in cols.items() if int(c) - int(r) >= 2)
    assert buggy == pytest.approx(any_neg)