"""Per-note RAG threshold gating before baseline mapping."""

from __future__ import annotations

from pathlib import Path

import pytest
from docx import Document

from backend.config import settings
from backend.core import ingest, section_mapper, template_discoverer
from backend.core.observation_matcher import (
    format_unmatched_observation_tag,
    observation_matches_baseline,
    partition_observations_for_baseline,
)
from backend.core.section_mapper import _format_unmatched_section_block


def _seed_roof_tenant(tmp_path: Path, monkeypatch, tenant: str = "obs-match") -> None:
    monkeypatch.setattr(settings, "data_dir", str(tmp_path / "data"))

    tpl = tmp_path / "template.docx"
    doc = Document()
    doc.add_paragraph("D2 Roof coverings")
    doc.save(str(tpl))

    past = tmp_path / "past.docx"
    doc2 = Document()
    doc2.add_paragraph("D2\nRoof coverings")
    doc2.add_paragraph(
        "The roof covering comprised slate tiles laid to pitched timber rafters. "
        "Inspection was limited to ground level using binoculars."
    )
    doc2.save(str(past))

    ingest.ingest_report_template(tenant, tpl)
    ingest.ingest_reference(tenant, past)


def test_format_unmatched_observation_tag():
    assert format_unmatched_observation_tag("asbestos noted") == (
        "[UNMATCHED_OBSERVATION: asbestos noted]"
    )


def test_unmatched_section_block_format():
    block = _format_unmatched_section_block(["slate slipped south slope", "moss on north"])
    assert "### UNMATCHED_OBSERVATION" in block
    assert "* slate slipped south slope" in block
    assert "* moss on north" in block


def test_partition_splits_low_match_notes(tmp_path, monkeypatch):
    tenant = "obs-part"
    _seed_roof_tenant(tmp_path, monkeypatch, tenant)
    schema = template_discoverer.load_schema(tenant)
    assert schema is not None
    baseline = (
        "The roof covering comprised slate tiles laid to pitched timber rafters. "
        "Inspection was limited to ground level using binoculars."
    )

    matched, unmatched = partition_observations_for_baseline(
        tenant,
        [
            "slate tile slipped south slope",
            "completely invented asbestos contamination in cavity",
        ],
        baseline,
        paragraph_section_id=schema.paragraph_section_id("D2"),
        interference_level="minimum",
    )
    assert matched
    assert any("slate" in m.lower() for m in matched)
    assert unmatched
    assert any("asbestos" in u.lower() for u in unmatched)


def test_unmatched_note_never_invented_in_section_text(tmp_path, monkeypatch):
    tenant = "obs-e2e"
    _seed_roof_tenant(tmp_path, monkeypatch, tenant)
    monkeypatch.setattr(settings, "note_rag_match_min_score", 0.38)

    result = section_mapper.generate_report(
        tenant,
        "slate tile slipped south slope\nasbestos contamination in roof void",
        interference_level="minimum",
    )
    d3 = next(s for s in result.sections if s.section_id == "D2")
    assert d3.status == "OK"
    mapped_body, _, _ = d3.text.partition("### UNMATCHED_OBSERVATION")
    assert "slate" in mapped_body.lower() or "slate" in d3.text.lower()
    assert "asbestos" not in mapped_body.lower()
    for sec in result.sections:
        if sec.section_id != "I3":
            assert "asbestos" not in (sec.text or "").lower()


def test_observation_matches_baseline_uses_lexical_overlap(tmp_path, monkeypatch):
    tenant = "obs-lex"
    _seed_roof_tenant(tmp_path, monkeypatch, tenant)
    baseline = "The chimney stacks were inspected from ground level using binoculars."
    assert observation_matches_baseline(
        tenant,
        "chimney stack hairline crack noted",
        baseline,
        paragraph_section_id="D1",
        interference_level="minimum",
    )