from __future__ import annotations import importlib.util from pathlib import Path import sys import numpy as np import pytest TOOLS = Path(__file__).parents[1] / "tools" def _load(name: str, filename: str): path = TOOLS / filename spec = importlib.util.spec_from_file_location(name, path) assert spec is not None and spec.loader is not None module = importlib.util.module_from_spec(spec) sys.modules[name] = module spec.loader.exec_module(module) return module runner = _load("h3_issue13_r2_gate_runner", "issue13_r2_gate_runner.py") analyzer = _load("h3_issue13_r2_analyze", "issue13_r2_analyze.py") def _source_prompt() -> dict: return { "305": { "class_type": "H3ContinuumSamplerV37", "inputs": { "continuation_backend": "Standard", "generation_mode": "Full Run", "review_action": "Continue / Next", "guide": ["900", 0], "last_frame": ["901", 0], }, }, "188": {"inputs": {"value": "old"}}, "153": {"inputs": {"scheduler": "old", "steps": 20, "denoise": 0.5}}, "297": {"inputs": {"value": 0.3}}, "306": {"inputs": {}}, "191": {"inputs": {}}, "249": {"inputs": {}}, "150": {"inputs": {"enabled": True}}, "286": { "inputs": { "lora_1": {"on": True, "lora": "turbo.safetensors", "strength": 1.0} } }, } def test_g1_configures_reported_six_by_ten_masked_av_baseline(): prompt = runner.configure_prompt( _source_prompt(), case="g1_standard_6x10", output_prefix="video/test", seed=130013, steps=8, size=384, diagnostic_nonce=7, ) sampler = prompt["305"] inputs = sampler["inputs"] assert sampler["class_type"] == "H3ContinuumMaskedPrefixR1Diagnostic" assert inputs["chunks"] == 6 assert inputs["chunk_seconds"] == 10.0 assert inputs["continuation_transport"] == "masked_av_prefix_22_v1" assert inputs["continuity"] == "Balanced — 22 frames" assert inputs["audio_continuity"] is True assert inputs["run_storage"] == "Off" assert "continuation_backend" not in inputs assert "guide" not in inputs assert "last_frame" not in inputs assert prompt["188"]["inputs"]["value"].count("\n---\n") == 5 assert prompt["150"]["inputs"]["enabled"] is False assert prompt["286"]["inputs"]["lora_1"]["on"] is False assert prompt["306"]["inputs"]["video_seam"] == "Off" assert prompt["306"]["inputs"]["audio_seam"] == "Off" def test_g2_keeps_old_five_by_five_comparison_shape(): prompt = runner.configure_prompt( _source_prompt(), case="g2_standard_5x5", output_prefix="video/test", seed=130013, steps=8, size=384, diagnostic_nonce=8, ) inputs = prompt["305"]["inputs"] assert inputs["chunks"] == 5 assert inputs["chunk_seconds"] == 5.0 assert prompt["188"]["inputs"]["value"].count("\n---\n") == 4 def test_diagnostic_payload_is_extracted_from_status_suffix(): assert runner._diagnostic_from_status( 'Production status\nV3.6-R1 diagnostic: {"transport":"masked_av_prefix_22_v1"}' ) == {"transport": "masked_av_prefix_22_v1"} def test_decoded_metrics_include_local_sharpness(): frame = np.zeros((32, 32, 3), dtype=np.uint8) frame[:, 16:] = 255 metrics = analyzer._frame_metrics(frame) assert metrics["contrast_p95_p05"] == 1.0 assert metrics["sharpness_laplacian_variance"] > 0.0 assert metrics["gradient_mean"] > 0.0 def test_latent_trend_requires_bit_exact_continuation_prefixes(): diagnostic = { "sample_calls": [ {"sample_number": 1, "output_tail": {"rms": 1.0, "std": 0.5}}, { "sample_number": 2, "source_tail": {"rms": 1.0, "std": 0.5}, "output_tail": {"rms": 1.2, "std": 0.7}, }, ], "final_prefix_pairs": [ { "bit_exact": True, "audio_bit_exact": True, "max_abs_diff": 0.0, "audio_max_abs_diff": 0.0, } ], } result = analyzer._latent_trend(diagnostic) assert result["all_video_prefix_pairs_bit_exact"] is True assert result["all_audio_prefix_pairs_bit_exact"] is True trend = result["streams"]["output_tail"]["trend"]["rms"] assert trend["last_minus_first"] == pytest.approx(0.2) assert trend["increase_steps"] == 1