tests/test_evaluation_panel_loader.py

"""Loader contract for app/components/evaluation_panel.py — issue #11 fix.

The Streamlit Evaluation tab globs ``data/evaluation/gold/*.json`` and tries
to parse each as ``GoldEvaluation``. Issue #11 added co-located
``*.diag.json`` files (NodeMatchCandidate dumps) that share the directory
but a different schema. Without filtering, the loader emits one
``st.warning`` per diag file with a noisy 11-error pydantic stack trace.

This test pins the filter so the bug doesn't regress.
"""

from __future__ import annotations

import json
from datetime import date
from pathlib import Path

from app.components.evaluation_panel import load_gold_evaluations
from src.models.schemas import (
    CascadeChain,
    CascadeNode,
    GoldEvaluation,
)


def _make_gold_evaluation_json(event_id: str) -> str:
    chain = CascadeChain(
        event_id=event_id,
        trigger_summary="trigger",
        trigger_country="C",
        trigger_iso="ISO",
        trigger_date=date(2025, 1, 1),
        cascade_events=[
            CascadeNode(
                id="E1",
                description="x",
                domain="infrastructure/power",
                severity="high",
                mechanism="m",
            )
        ],
    )
    ge = GoldEvaluation(
        event_id=event_id,
        predicted_chain=chain,
        gold_event_id=event_id,
        gold_node_count=1,
        predicted_node_count=1,
        matches=[],
        precision=0.0,
        recall=0.0,
        f1=0.0,
        domain_jaccard=0.0,
        threshold=0.35,
        embedder_model="all-MiniLM-L6-v2",
        algorithm_version="v3",
        gold_eval_fingerprint="abc123",
        evaluated_at=date(2025, 1, 1),
    )
    return ge.model_dump_json(indent=2)


def _diag_payload(event_id: str) -> str:
    return json.dumps({
        "event_id": event_id,
        "threshold_used": 0.5,
        "embedder_model": "all-MiniLM-L6-v2",
        "algorithm_version": "v3",
        "candidates": [
            {"p_id": "E1", "g_id": "G1", "cosine": 0.42,
             "severity_match": True, "accepted": False}
        ],
    })


def test_loader_skips_diag_json_files(tmp_path: Path):
    """A directory with both .json and .diag.json should yield only the
    GoldEvaluation files — diag files are silently skipped, no warning."""
    (tmp_path / "2025-0183-ITA.json").write_text(
        _make_gold_evaluation_json("2025-0183-ITA"), encoding="utf-8"
    )
    (tmp_path / "2025-0183-ITA.diag.json").write_text(
        _diag_payload("2025-0183-ITA"), encoding="utf-8"
    )
    (tmp_path / "2025-0848-UKR.json").write_text(
        _make_gold_evaluation_json("2025-0848-UKR"), encoding="utf-8"
    )
    (tmp_path / "2025-0848-UKR.diag.json").write_text(
        _diag_payload("2025-0848-UKR"), encoding="utf-8"
    )

    evaluations = load_gold_evaluations(tmp_path)
    assert len(evaluations) == 2
    assert {e.event_id for e in evaluations} == {"2025-0183-ITA", "2025-0848-UKR"}


def test_loader_returns_empty_when_dir_missing(tmp_path: Path):
    assert load_gold_evaluations(tmp_path / "nope") == []


def test_loader_handles_diag_only_dir(tmp_path: Path):
    """Pathological case: only .diag.json files exist. Should return [],
    not crash, not emit warnings about diag files."""
    (tmp_path / "2025-0183-ITA.diag.json").write_text(
        _diag_payload("2025-0183-ITA"), encoding="utf-8"
    )
    assert load_gold_evaluations(tmp_path) == []


def test_loader_skips_outlier_event_ids(tmp_path: Path):
    """Issue #14 fix: even with a stale GoldEvaluation file on disk for
    an event that was later marked as outlier in config.evaluation.
    outlier_event_ids, the loader must hide it so the Streamlit
    Evaluation tab matches what scripts/05_evaluate.py aggregates."""
    (tmp_path / "2025-0183-ITA.json").write_text(
        _make_gold_evaluation_json("2025-0183-ITA"), encoding="utf-8"
    )
    (tmp_path / "2025-0632-ROU.json").write_text(
        _make_gold_evaluation_json("2025-0632-ROU"), encoding="utf-8"
    )
    (tmp_path / "2025-0848-UKR.json").write_text(
        _make_gold_evaluation_json("2025-0848-UKR"), encoding="utf-8"
    )

    # No outlier set -> all three returned (back-compat baseline).
    assert {e.event_id for e in load_gold_evaluations(tmp_path)} == {
        "2025-0183-ITA", "2025-0632-ROU", "2025-0848-UKR",
    }

    # ROU marked outlier -> hidden even though the file still sits on disk.
    filtered = load_gold_evaluations(
        tmp_path, outlier_event_ids={"2025-0632-ROU"}
    )
    assert {e.event_id for e in filtered} == {"2025-0183-ITA", "2025-0848-UKR"}