tests/test_engine.py

from __future__ import annotations

from brain.decision_maker import choose_mail_action
from core_engine.evaluator import TriageEvaluator
from core_engine.schemas import AgentDecision, SyntheticMail
from core_engine.simulator import SimulationEngine


def test_reset_generates_emails():
    engine = SimulationEngine(batch_size=3, random_seed=11)

    state = engine.reset()

    assert len(state["emails"]) == 3
    assert state["progress"] == {"processed": 0, "remaining": 3, "total": 3}
    assert state["current_email"]["id"] == state["emails"][0]["id"]


def test_step_updates_state_and_score():
    engine = SimulationEngine(batch_size=2, random_seed=13)
    state = engine.reset()
    decision = choose_mail_action(state["current_email"])

    result = engine.step(decision)

    assert result["done"] is False
    assert result["state"]["progress"]["processed"] == 1
    assert 0 < result["reward"] < 1
    assert 0 < result["score"]["classification_accuracy"] < 1


def test_evaluator_returns_accuracy_scores_between_zero_and_one():
    evaluator = TriageEvaluator()
    message = SyntheticMail(
        mail_id="mail_test",
        sender="ops@example.com",
        subject="Urgent production review needed",
        body="Please review asap.",
        truth_category="urgent",
    )
    decision = AgentDecision(
        mail_id="mail_test",
        predicted_category="urgent",
        priority_level="high",
    )

    record = evaluator.evaluate(message, decision)
    summary = evaluator.summarize([record], total_count=1)

    assert 0 < summary.classification_accuracy < 1
    assert 0 < summary.priority_correctness < 1
    assert 0 < summary.weighted_score < 1
    assert summary.numeric_score == 0.999999
    assert summary.confusion_matrix["urgent"]["urgent"] == 1


def test_urgent_misclassification_gets_penalty():
    evaluator = TriageEvaluator()
    message = SyntheticMail(
        mail_id="mail_urgent",
        sender="ops@example.com",
        subject="Urgent production review needed",
        body="Please review asap.",
        truth_category="urgent",
    )
    decision = AgentDecision(
        mail_id="mail_urgent",
        predicted_category="general",
        priority_level="low",
    )

    record = evaluator.evaluate(message, decision)
    summary = evaluator.summarize([record], total_count=1)

    assert record.urgent_penalty_applied is True
    assert summary.urgent_penalty_count == 1
    assert summary.weighted_score == 0.000001


def test_all_wrong_score_is_strictly_above_zero():
    evaluator = TriageEvaluator()
    message = SyntheticMail(
        mail_id="mail_wrong",
        sender="ops@example.com",
        subject="Urgent production review needed",
        body="Please review asap.",
        truth_category="urgent",
    )
    decision = AgentDecision(
        mail_id="mail_wrong",
        predicted_category="general",
        priority_level="low",
    )

    summary = evaluator.summarize([evaluator.evaluate(message, decision)], total_count=1)

    assert 0 < summary.numeric_score < 1


def test_all_correct_score_is_strictly_below_one():
    evaluator = TriageEvaluator()
    message = SyntheticMail(
        mail_id="mail_correct",
        sender="shop@example.com",
        subject="Weekend sale",
        body="Discount available today.",
        truth_category="promotion",
    )
    decision = AgentDecision(
        mail_id="mail_correct",
        predicted_category="promotion",
        priority_level="medium",
    )

    summary = evaluator.summarize([evaluator.evaluate(message, decision)], total_count=1)

    assert 0 < summary.numeric_score < 1


def test_normal_non_boundary_score_is_unchanged():
    evaluator = TriageEvaluator()
    message = SyntheticMail(
        mail_id="mail_partial",
        sender="shop@example.com",
        subject="Weekend sale",
        body="Discount available today.",
        truth_category="promotion",
    )
    decision = AgentDecision(
        mail_id="mail_partial",
        predicted_category="general",
        priority_level="medium",
    )

    summary = evaluator.summarize([evaluator.evaluate(message, decision)], total_count=1)

    assert summary.numeric_score == 0.25