tests/test_label_leak.py

"""CI gate: a lesion label/mask must NEVER be accessible during subspace construction.

This test must FAIL the build if any label read can occur inside
`subspace_construction_guard()` (IMPLEMENTATION_SPEC §5, CLAUDE.md). It exercises the
runtime guard directly and also simulates a subspace `fit` that wrongly tries to read a
label, asserting the guard stops it.
"""
import sys
from pathlib import Path

import pytest

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from data.leak_guard import (  # noqa: E402
    LabelLeakError,
    assert_label_free,
    subspace_construction_guard,
)
from data.masks import label_from_manifest_record  # noqa: E402

_REC = {
    "slice_id": "s0",
    "has_nodule": True,
    "nodule_pixel_area": 42.0,
    "nodule_diameter_mm": 8.1,
    "label": "tumor",
}


def test_label_read_allowed_outside_construction():
    # EVAL paths (outside subspace construction) may read labels freely.
    lab = label_from_manifest_record(_REC)
    assert lab.has_nodule is True
    assert lab.label == "tumor"


def test_label_read_blocked_during_construction():
    with subspace_construction_guard():
        with pytest.raises(LabelLeakError):
            label_from_manifest_record(_REC)


def test_raw_guard_raises_inside_construction():
    with subspace_construction_guard():
        with pytest.raises(LabelLeakError):
            assert_label_free("lesion mask")


def test_guard_is_reentrant_and_restores():
    with subspace_construction_guard():
        with subspace_construction_guard():
            with pytest.raises(LabelLeakError):
                assert_label_free()
    # outside again: no raise
    assert_label_free()


def test_simulated_subspace_fit_with_leak_fails():
    """A subspace fit() that illegally reads a label must blow up via the guard."""

    def bad_fit(token_bank, record):
        with subspace_construction_guard():
            # ILLEGAL: peeking at the lesion label to shape the subspace.
            _ = label_from_manifest_record(record)
            return token_bank.mean()

    with pytest.raises(LabelLeakError):
        bad_fit([1.0, 2.0], _REC)