chore: add remaining test stubs and script placeholders

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/bench_mlx.py

@@ -0,0 +1,9 @@
+#!/usr/bin/env python3
+"""Benchmark MLX inference on Apple Silicon.
+
+Reports latency, throughput, and memory usage.
+
+Usage:
+    uv run python scripts/bench_mlx.py
+"""
+# TODO: Phase 5 — implement after full pipeline works

scripts/compare_reference.py

@@ -0,0 +1,9 @@
+#!/usr/bin/env python3
+"""Compare MLX outputs against PyTorch reference fixtures.
+
+Reports max abs error and mean abs error per tensor.
+
+Usage:
+    uv run python scripts/compare_reference.py
+"""
+# TODO: Phase 2+ — implement after reference fixtures exist

tests/test_conversion.py

@@ -0,0 +1,23 @@
+"""Tests: weight conversion PyTorch → MLX (Phase 3).
+
+Validates key mapping, shape transforms, and round-trip integrity.
+"""
+
+import pytest
+
+pytestmark = pytest.mark.skip(reason="Phase 3: converter not yet implemented")
+
+
+def test_key_mapping_complete() -> None:
+    """Every PyTorch key maps to an MLX key with no orphans."""
+    ...
+
+
+def test_conv_weight_transpose() -> None:
+    """Conv weights correctly transposed from NCHW → NHWC."""
+    ...
+
+
+def test_first_conv_4ch_preserved() -> None:
+    """Patched 4-channel first conv preserved during conversion."""
+    ...

tests/test_decoder_parity.py

@@ -0,0 +1,19 @@
+"""Parity tests: MLX decoder heads vs PyTorch reference (Phase 2).
+
+Uses saved backbone features → runs MLX decoder → compares
+against saved PyTorch coarse predictions.
+"""
+
+import pytest
+
+pytestmark = pytest.mark.skip(reason="Phase 2: MLX decoder not yet implemented")
+
+
+def test_alpha_decoder_parity() -> None:
+    """MLX alpha decoder matches PyTorch within tolerance."""
+    ...
+
+
+def test_fg_decoder_parity() -> None:
+    """MLX foreground decoder matches PyTorch within tolerance."""
+    ...

tests/test_import.py

@@ -0,0 +1,12 @@
+"""Smoke test: verify package imports succeed."""
+
+
+def test_package_imports() -> None:
+    import corridorkey_mlx
+    import corridorkey_mlx.convert
+    import corridorkey_mlx.inference
+    import corridorkey_mlx.io
+    import corridorkey_mlx.model
+    import corridorkey_mlx.utils
+
+    assert corridorkey_mlx.__version__ == "0.1.0"

tests/test_refiner_parity.py

@@ -0,0 +1,19 @@
+"""Parity tests: MLX refiner vs PyTorch reference (Phase 2).
+
+Uses saved coarse predictions + RGB → runs MLX refiner →
+compares against saved PyTorch final outputs.
+"""
+
+import pytest
+
+pytestmark = pytest.mark.skip(reason="Phase 2: MLX refiner not yet implemented")
+
+
+def test_refiner_delta_parity() -> None:
+    """MLX refiner delta logits match PyTorch within tolerance."""
+    ...
+
+
+def test_refiner_final_output_parity() -> None:
+    """MLX final alpha and fg match PyTorch within tolerance."""
+    ...