tests/test_engine.py

"""Integration tests for the ExperimentationOS orchestrator."""

from __future__ import annotations

import pytest

from spinor_os.config import AttributionMethod, Direction, EventType, LoopStage
from spinor_os.engine import ExperimentationOS
from spinor_os.workflow import WorkflowTransitionError


def make_os() -> ExperimentationOS:
    os = ExperimentationOS(config={"min_replications": 1})
    os.register_employee("emp-001", "rep", "northeast", {"enterprise": 0.8})
    return os


def test_health_check_empty():
    os = ExperimentationOS()
    health = os.health_check()
    assert health["status"] == "healthy"


def test_full_experiment_loop():
    os = make_os()
    hypothesis = os.propose_hypothesis(
        statement="X increases Y",
        causal_claim="X causes Y to increase.",
        predicted_effect={
            "metric": "y_rate",
            "direction": Direction.INCREASE.value,
            "magnitude": 0.1,
            "unit": "pp",
            "timing": "7d",
        },
        employee_owner="emp-001",
        falsification_criteria=["Y does not increase"],
        customer_segment="enterprise",
        territory="northeast",
        modification="mod-a",
    )

    exp = os.start_experiment(hypothesis.hypothesis_id)
    assert exp.current_stage == LoopStage.RESEARCH

    candidates = [
        {
            "hypothesis_id": hypothesis.hypothesis_id,
            "modification": "mod-a",
            "customer_segment": "enterprise",
            "territory": "northeast",
            "timing": "2026-W31",
            "resource_allocation": 1.0,
        }
    ]
    mission = os.allocate_mission(exp.experiment_id, "emp-001", candidates)
    assert mission.mission_id in os.missions

    os.record_event(
        exp.experiment_id,
        mission.mission_id,
        EventType.MISSION_EXECUTED,
        actor_id="emp-001",
        execution_quality=0.85,
    )
    os.record_event(
        exp.experiment_id,
        mission.mission_id,
        EventType.OUTCOME_OBSERVED,
        actor_id="emp-001",
        outcome_value=0.2,
        metric="y_rate",
        execution_quality=0.85,
    )

    claim = os.attribute(
        exp.experiment_id,
        outcome_metric="y_rate",
        outcome_value=0.2,
        counterfactual_estimate=0.05,
        method=AttributionMethod.RCT,
        confidence=0.95,
        falsification_survived=True,
    )
    assert claim.is_significant()

    os.advance(exp.experiment_id, LoopStage.OBSERVE)
    os.advance(exp.experiment_id, LoopStage.ATTRIBUTE)
    os.advance(exp.experiment_id, LoopStage.FALSIFY)
    assert exp.current_stage == LoopStage.FALSIFY

    os.ingest_to_causal_graph(exp.experiment_id)
    os.update_policy(exp.experiment_id)

    # Engine can produce top causal graph candidates.
    top = os.causal_graph.next_experiments()
    assert len(top) == 1


def test_promotion_requires_replications():
    os = make_os()
    hypothesis = os.propose_hypothesis(
        statement="X increases Y",
        causal_claim="X causes Y to increase.",
        predicted_effect={
            "metric": "y_rate",
            "direction": Direction.INCREASE.value,
            "magnitude": 0.1,
            "unit": "pp",
            "timing": "7d",
        },
        employee_owner="emp-001",
        falsification_criteria=["Y does not increase"],
        customer_segment="enterprise",
        territory="northeast",
        modification="mod-a",
    )
    exp = os.start_experiment(hypothesis.hypothesis_id)
    candidates = [
        {
            "hypothesis_id": hypothesis.hypothesis_id,
            "modification": "mod-a",
            "customer_segment": "enterprise",
            "territory": "northeast",
            "timing": "2026-W31",
            "resource_allocation": 1.0,
        }
    ]
    mission = os.allocate_mission(exp.experiment_id, "emp-001", candidates)
    os.record_event(
        exp.experiment_id,
        mission.mission_id,
        EventType.OUTCOME_OBSERVED,
        actor_id="emp-001",
        outcome_value=0.2,
        metric="y_rate",
        execution_quality=0.85,
    )
    os.attribute(
        exp.experiment_id,
        outcome_metric="y_rate",
        outcome_value=0.2,
        counterfactual_estimate=0.05,
        method=AttributionMethod.RCT,
        confidence=0.95,
        falsification_survived=True,
    )
    os.advance(exp.experiment_id, LoopStage.OBSERVE)
    os.advance(exp.experiment_id, LoopStage.ATTRIBUTE)
    os.advance(exp.experiment_id, LoopStage.FALSIFY)
    os.advance(exp.experiment_id, LoopStage.REPLICATE)
    os.advance(exp.experiment_id, LoopStage.SYSTEMIZE)

    strategy = os.systemize_strategy(exp.experiment_id, "S", "D", automation_ready=True)
    # Promotion should fail because the registry requires 3 replications and this experiment has only 1.
    os.golden_registry.rules.required_replications = 3
    os.golden_registry.rules.required_adversarial_retests = 0
    result = os.promote_to_golden(strategy.strategy_id, "Mechanism", "auto-1")
    assert not result.success
    assert any("replications" in r.lower() for r in result.reasons)