tests/test_methods_comparison.py

"""Confronto "metodi" DERIVATO dalle tracce della singola run (debug == analizza).

Il tab metodi non ri-esegue i layer: ricostruisce "chi ha catturato cosa" dai
contributi grezzi già presenti nelle tracce della run di analisi.
"""

from __future__ import annotations

from core.output.trace import ClusterTrace, SpanContribution
from web.services.analysis import build_methods_comparison

# process -> (layer_key, label)
PROCESS_INDEX = {
    "gliner": ("zero_shot", "Zero-shot (GLiNER)"),
    "rule_cig": ("rules", "Regole deterministiche"),
}


def _contrib(process, text, etype, score):
    return SpanContribution(
        process=process, layer="X", entity_type=etype, text=text,
        start=0, end=len(text), score=score, validated=False,
        source=process, is_representative=False,
    )


def _trace(contribs):
    return ClusterTrace(
        text=contribs[0].text, entity_type=contribs[0].entity_type, label="",
        severity="", start=0, end=10, base_score=0.5, agreement_boost=0.0,
        final_score=0.5, agreement_count=1, validated=False, threshold=0.5,
        passed_threshold=True, kept=True, contributions=contribs,
    )


def test_only_selected_layers_appear():
    """Selezionato solo zero_shot: un contributo 'rules' NON deve comparire."""
    traces = [_trace([
        _contrib("gliner", "CIG1234", "CIG", 0.9),
        _contrib("rule_cig", "CIG1234", "CIG", 1.0),
    ])]
    out = build_methods_comparison(traces, PROCESS_INDEX, selected={"zero_shot"})

    assert out["methods"] == ["zero_shot"]
    assert "rules" not in out["method_labels"]
    assert out["entities"][0]["found_by"] == ["zero_shot"]


def test_found_by_and_not_found_by_across_layers():
    """Entrambi i layer selezionati; entità catturata solo da uno."""
    traces = [_trace([_contrib("gliner", "Mario Rossi", "RUP", 0.8)])]
    out = build_methods_comparison(
        traces, PROCESS_INDEX, selected={"zero_shot", "rules"}
    )
    e = out["entities"][0]
    assert e["found_by"] == ["zero_shot"]
    assert e["not_found_by"] == ["rules"]
    assert e["by_method"]["zero_shot"]["score"] == 0.8
    assert e["by_method"]["zero_shot"]["processes"] == ["gliner"]


def test_scores_use_best_contribution_per_layer():
    traces = [
        _trace([_contrib("gliner", "Comune di Milano", "STAZIONE_APPALTANTE", 0.7)]),
        _trace([_contrib("gliner", "Comune di Milano", "STAZIONE_APPALTANTE", 0.95)]),
    ]
    out = build_methods_comparison(traces, PROCESS_INDEX, selected={"zero_shot"})
    e = out["entities"][0]
    assert e["by_method"]["zero_shot"]["score"] == 0.95
    assert e["by_method"]["zero_shot"]["count"] == 2