# =============================================================================
#  Project   : modelmap
#  File      : tests/test_stats_and_probes.py
#  Purpose   : Correctness tests for stats + probe modules (tiny references)
#  Author    : Simon-Pierre Boucher
#  Contact   : contact@spboucher.ai
#  Website   : https://modelmap.io
#  Created   : 2026-08-12
#  Modified  : 2026-08-12
#  Platform  : macOS / Apple Silicon (arm64)
#  License   : All rights reserved (research code)
# =============================================================================
"""Charter §0.3: numerical probes get correctness tests against tiny
references — a probe must separate real structure from noise, FDR must
control false discoveries on known nulls, map cards must gate publication."""

import numpy as np

from modelmap.atlas.mapcard import MapCard
from modelmap.probes.linear import probe_with_control
from modelmap.stats.replication import bh_fdr, bootstrap_ci, jaccard, replication_rate


def test_bootstrap_ci_covers_mean():
    rng = np.random.default_rng(1)
    point, lo, hi = bootstrap_ci(rng.normal(5.0, 1.0, 500))
    assert lo < 5.0 < hi and abs(point - 5.0) < 0.2


def test_bh_fdr_null_and_signal():
    rng = np.random.default_rng(2)
    null_p = rng.uniform(0, 1, 1000)
    assert bh_fdr(null_p, q=0.05).mean() < 0.01  # near-zero false discoveries
    mixed = np.concatenate([null_p, np.full(50, 1e-8)])
    disc = bh_fdr(mixed, q=0.05)
    assert disc[-50:].all()  # all planted signals found


def test_jaccard_and_replication():
    assert jaccard({1, 2}, {1, 2}) == 1.0
    assert jaccard({1}, {2}) == 0.0
    point, lo, hi = replication_rate([{1, 2, 3}, {2, 3, 4}, {1, 2, 3}])
    assert 0 < lo <= point <= hi <= 1


def test_probe_separates_structure_from_noise():
    """On separable data: high selectivity. On pure noise labels: ~0."""
    rng = np.random.default_rng(3)
    n, d = 600, 32
    y = rng.integers(0, 2, n)
    x = rng.normal(0, 1, (n, d))
    x[:, 0] += 3.0 * y  # planted signal in one dimension
    r = probe_with_control(x, y, seed=0)
    assert r.task_accuracy > 0.9
    assert r.selectivity > 0.3

    y_noise = rng.integers(0, 2, n)
    x_noise = rng.normal(0, 1, (n, d))
    r0 = probe_with_control(x_noise, y_noise, seed=0)
    assert abs(r0.selectivity) < 0.15  # noise floor: no fake structure


def test_mapcard_gates_publication():
    card = MapCard()
    assert card.validate()  # empty card blocked
    card = MapCard(
        map_id="atlas/demo/probes/v0", map_type="probes", model_id="demo",
        model_hash="abc", quantization="fp16", commit="deadbeef",
        config="{}", created="2026-08-12", hardware_manifest={"chip": "test"},
        confidence_level=1, regenerate_command="make demo",
        seeds=[0, 1, 2], prompt_sets=["a#sha", "b#sha"],
        controls=["shuffled-labels"], replication_rate=0.8,
    )
    assert card.validate() == []
    card.seeds = [0]  # Level 1 with 1 seed must fail
    assert any("3 seeds" in e for e in card.validate())