"""Explicit opt-in acceptance of real profiles and configured model API.

This is DEV/regression coverage, not an independent estimate of accuracy.
RUN_KNOWLEDGE_LIVE=1 is required; the normal suite never sends wiki texts out.
"""

import json
import re
import os
from pathlib import Path

import pytest

ROOT = Path(__file__).resolve().parents[1]
CASES = json.loads((ROOT / "tools/knowledge/questions.json").read_text())["questions"]
pytestmark = pytest.mark.skipif(
    os.getenv("RUN_KNOWLEDGE_LIVE") != "1", reason="Explicit permission and model API required"
)


@pytest.mark.parametrize("engine", ["rag", "graphrag"])
@pytest.mark.parametrize("case", CASES, ids=lambda c: c["id"])
def test_real_profile_answer_preserves_evidence(case, engine, monkeypatch):
    if (
        case["id"] == "content-17"
        and engine != "rag"
        or case["id"] == "content-18"
        and engine != "graphrag"
    ):
        pytest.skip("Metadata regression belongs to its reference engine")
    from dotenv import dotenv_values

    allowed = {
        "OPENWEBUI_API_KEY",
        "ANSWER_PROVIDER",
        "ANSWER_MODEL",
        "ANSWER_LLM_URL",
        "ANSWER_API_KEY",
        "ANTHROPIC_API_KEY",
        "NEO4J_URI",
        "NEO4J_USER",
        "NEO4J_PASSWORD",
        "NEO4J_DATABASE",
    }
    for key, value in dotenv_values(ROOT / ".env").items():
        if key in allowed and value is not None:
            monkeypatch.setenv(key, value)
    from tools.answers.answer_delivery import deliver_answer
    from tools.answers.context_policy import normalized

    result = deliver_answer(case["query"], engine=engine, limit=10)
    assert result["accepted_for_delivery"] and result["status"] != "error", result.get("reason")
    if "expected_status" in case:
        assert result["status"] == case["expected_status"]
    for text in case.get("answer_contains", []):
        assert text in result["answer"]
    for text in case.get("answer_excludes", []):
        assert text not in result["answer"]
    evidence = result.get("evidence", [])
    for row in evidence:
        source = ROOT.parent / "zpwiki" / row["source_path"]
        assert row["text"] in source.read_text(), row["id"]
        assert row["source_url"] in result["source_urls"]
    if case.get("expected_person"):
        assert any(row["person"] == case["expected_person"] for row in evidence), result["answer"]
    if case.get("expected_date"):
        assert any(
            row["date"] == case["expected_date"] and case["evidence_contains"] in row["text"]
            for row in evidence
        )
    if case.get("missing_results"):
        assert "nedokladá" in result["answer"]
        assert not any(row["phase"] == "result" and row["person"] == "Ján Pták" for row in evidence)
    if case.get("requires_both_sides"):
        people = {r["person"] for r in evidence}
        assert "Ján Pták" in people and people & set(case["expected_any_related_people"])
        # A bounded discovery query does not prescribe one ranking. Require
        # source proof of the research topic, not only a shared UI or webhook.
        candidate_text = " ".join(r["text"] for r in evidence if r["person"] != "Ján Pták")
        assert re.search(
            r"\brag\b|\b(?:graph|light)rag\b|\bznalostn\w* graf\w*\b|"
            r"\bgrafov\w* vyhladav\w*\b|\bknowledge graph\b",
            normalized(candidate_text),
        )
    if case.get("must_disclaim_current_activity"):
        assert "nepotvrdzuje" in result["answer"] and "aktívne" in result["answer"]
    if case.get("requires_explicit_meeting_date"):
        assert evidence and all(
            row["date"] and "2020-01-01" <= row["date"] <= "2021-12-31" for row in evidence
        )
    if case.get("requires_recognized_name_notice"):
        assert "Rozpoznané meno" in result["answer"]
    if case["id"] == "content-07" and result["status"] == "ok":
        # Literal titles of the same work are supporting context, not claims
        # that the title itself proves annotation was actually performed.
        primary = [r for r in evidence if r.get("evidence_role") != "work_context"]
        assert primary and all(
            "anot" in normalized(row["text"])
            and ("ner" in normalized(row["text"]).split() or "entit" in normalized(row["text"]))
            for row in primary
        )
    if case["id"] == "content-17":
        assert (
            "Systém získavania informácií v slovenskom jazyku" in result["answer"]
            and "2021" in result["answer"]
        )
    if case["id"] == "content-18":
        assert "NLP" in result["answer"] and "RAG" in result["answer"]