"""
TradeSense AI — Multiple-Testing & Experiment Registry Unit Tests
Validates Benjamini-Hochberg FDR, Holm-Bonferroni correction,
pre-registered hypothesis tracking, and promotion status adjustments.
"""

from datetime import datetime, timezone
import pytest
import numpy as np

from quant.evaluation.experiment_registry import (
    ExperimentRegistry,
    MultipleTestingCorrector,
    MultipleTestingMethod,
    HypothesisType,
    ExperimentHypothesis,
    ExperimentRecord,
)


class TestMultipleTestingCorrections:
    def test_benjamini_hochberg_ordering_and_adjustment(self):
        """
        BH adjustment must produce p_adj >= p_raw and preserve ranking.
        """
        raw_p = [0.001, 0.01, 0.04, 0.10, 0.50]
        adj_p, sigs = MultipleTestingCorrector.benjamini_hochberg(raw_p, q_threshold=0.05)

        for p_r, p_a in zip(raw_p, adj_p):
            assert p_a >= p_r

        assert adj_p[0] < adj_p[1] < adj_p[2] <= adj_p[3] <= adj_p[4]
        # First 3 should remain significant at FDR 0.05
        assert sigs[0] is True
        assert sigs[1] is True
        assert sigs[4] is False

    def test_holm_bonferroni_step_down(self):
        """
        Holm-Bonferroni step down tests: p_adj_1 = 5 * 0.005 = 0.025.
        """
        raw_p = [0.005, 0.02, 0.03, 0.08, 0.20]
        adj_p, sigs = MultipleTestingCorrector.holm_bonferroni(raw_p, alpha=0.05)

        assert adj_p[0] == pytest.approx(0.025, abs=1e-4)
        assert sigs[0] is True
        assert all(adj_p[i] >= raw_p[i] for i in range(len(raw_p)))


class TestExperimentRegistry:
    def test_pre_registered_hypothesis_locking(self):
        registry = ExperimentRegistry()
        hypo = registry.register_hypothesis(
            trial_family_id="fam_test_001",
            hypothesis_statement="Candidate model outperforms Naive persistence on 21D horizon.",
            hypothesis_type=HypothesisType.CONFIRMATORY,
        )

        assert hypo.is_locked is True
        assert hypo.hypothesis_id.startswith("hypo_fam_test_001_")
        assert hypo.hypothesis_type == HypothesisType.CONFIRMATORY

    def test_multiple_testing_downgrades_fragile_promotion(self):
        """
        If a model is promoted on raw p=0.04, but after testing 10 models
        its adjusted p becomes 0.20, the registry must downgrade it to
        NO_ACTIONABLE_EDGE with explicit rejection reason.
        """
        registry = ExperimentRegistry()
        hypo = registry.register_hypothesis(
            trial_family_id="fam_downgrade_test",
            hypothesis_statement="Confirmatory multi-model test.",
        )

        # Record 1: Candidate with borderline raw p=0.04
        registry.record_experiment_result(
            hypothesis=hypo,
            ticker="AAPL",
            horizon_sessions=21,
            dataset_lineage_hash="hash_aapl",
            feature_version="fts_v1",
            model_version="xgboost_v1",
            hyperparameter_version="hp_v1",
            promotion_policy_version="prom_v1",
            random_seed=42,
            raw_dm_p_value=0.04,  # Borderline significant before correction
            oos_rmse_improvement_pct=15.0,
            directional_accuracy_pct=58.0,
            mean_calibration_error_pct=2.0,
            gross_return_pct=18.0,
            net_return_pct=14.0,
            net_sharpe=1.5,
            max_drawdown_pct=10.0,
            is_robust=True,
            initial_promotion_status="PROMOTED",
            initial_signal_decision="BULLISH",
            initial_rejection_reasons=[],
        )

        # Add 9 other non-significant trials to inflate the family size
        for i in range(2, 11):
            registry.record_experiment_result(
                hypothesis=hypo,
                ticker=f"STOCK_{i}",
                horizon_sessions=21,
                dataset_lineage_hash=f"hash_{i}",
                feature_version="fts_v1",
                model_version="xgboost_v1",
                hyperparameter_version="hp_v1",
                promotion_policy_version="prom_v1",
                random_seed=42,
                raw_dm_p_value=0.30 + (i * 0.05),
                oos_rmse_improvement_pct=2.0,
                directional_accuracy_pct=50.0,
                mean_calibration_error_pct=4.0,
                gross_return_pct=2.0,
                net_return_pct=-1.0,
                net_sharpe=0.1,
                max_drawdown_pct=18.0,
                is_robust=True,
                initial_promotion_status="REJECTED",
                initial_signal_decision="NO_ACTIONABLE_EDGE",
                initial_rejection_reasons=["INSUFFICIENT_RMSE_EDGE"],
            )

        # Apply Benjamini-Hochberg
        corrected = registry.apply_multiple_testing_correction(
            trial_family_id="fam_downgrade_test",
            method=MultipleTestingMethod.BENJAMINI_HOCHBERG,
            alpha_fdr=0.05,
        )

        aapl_rec = [r for r in corrected if r.ticker == "AAPL"][0]
        # In family of 10, raw p=0.04 adjusted is min(1.0, 10/1 * 0.04) = 0.40 > 0.05
        assert aapl_rec.adjusted_dm_p_value > 0.05
        assert aapl_rec.promotion_status == "REJECTED"
        assert aapl_rec.signal_decision == "NO_ACTIONABLE_EDGE"
        assert any("MULTIPLE_TESTING_FAIL" in r for r in aapl_rec.rejection_reasons)