tests/test_eval.py

from __future__ import annotations

import json
from pathlib import Path

from bl_ocr.domain.schema_validation import build_validator
from bl_ocr.eval.harness import evaluate_annotation, evaluate_directory

PROJECT_ROOT = Path(__file__).resolve().parents[1]
SCHEMA_PATH = PROJECT_ROOT / "schema.json"


def _validator():
    return build_validator(json.loads(SCHEMA_PATH.read_text(encoding="utf-8")))


def _example() -> dict:
    return json.loads((PROJECT_ROOT / "example.json").read_text(encoding="utf-8"))


def test_clean_example_passes_all_structural_checks():
    result = evaluate_annotation("example", _example(), _validator())
    assert result.parse_pass
    assert result.schema_pass
    assert result.schema_error_count == 0
    assert result.row_count_pass
    assert result.arithmetic_pass
    assert not result.critical


def test_row_count_mismatch_is_flagged():
    annotation = _example()
    annotation["totals"]["line_count"]["value"] = 99
    result = evaluate_annotation("mismatch", annotation, _validator())
    assert not result.row_count_pass


def test_empty_products_with_line_count_is_critical():
    annotation = _example()
    annotation["products"] = []
    result = evaluate_annotation("empty", annotation, _validator())
    assert result.critical
    assert result.critical_reason is not None


def test_none_annotation_fails_parse():
    result = evaluate_annotation("broken", None, _validator(), parse_error="bad json")
    assert not result.parse_pass
    assert result.parse_error == "bad json"


def test_evaluate_directory_aggregates(tmp_path: Path):
    # one clean file, one schema-invalid file (extra key)
    clean = _example()
    (tmp_path / "clean.json").write_text(
        json.dumps({"document_annotation": json.dumps(clean)}), encoding="utf-8"
    )
    dirty = _example()
    dirty["products"][0]["_repaired"] = "leak"
    (tmp_path / "dirty.json").write_text(
        json.dumps({"document_annotation": json.dumps(dirty)}), encoding="utf-8"
    )

    summary = evaluate_directory(tmp_path, SCHEMA_PATH)
    assert summary.total == 2
    assert summary.parse_pass == 2
    assert summary.schema_pass == 1
    data = summary.as_dict()
    assert data["rates"]["schema_pass"] == 0.5
    assert len(data["files"]) == 2