PROTOTYPE_STATUS.md

Prototype Status Matrix

Generated for Task 11 (prototype sweep). Status is determined by the automated smoke matrix in tests/test_prototype_smoke.py (27 legacy prototypes, construction-only bar for manual-run() classes, bounded update-loop for PrototypeBase subclasses) plus the dedicated tests for audio_mirror.py and ai-mirror-of-truth.py. Run with:

LUNAR_HEADLESS=1 python3 -m pytest tests/test_prototype_smoke.py -v

Legend:

• works — instantiates and runs cleanly against the headless tools layer; no known functional gaps blocking use.
• degraded — instantiates and runs, but a code path is known to silently no-op / fall back to a stub instead of failing (see reason).
• needs-rework — fails the smoke bar; requires non-trivial changes beyond this sweep's scope.

verified-on-hardware reflects whether the prototype has been run against real hardware (mic/webcam/renderer/afterwords server) since the MLX-native rework began; this sweep only validates the headless path, so it is no for everything unless otherwise noted.

Prototype	Status	Reason	Verified on hardware
acoustic-fingerprint-painter.py	works (fixed)	Fixed in this sweep: _get_stroke_parameters_from_gpt4 now guards a None LLM backend and None/empty responses, parses via a fence/JSON-block-tolerant helper, and _draw_stroke sanitizes numeric/RGB inputs; the headless FakeAudioRecorder writes a real WAV so feature extraction succeeds.	no
ai-dream-interpreter-prototype.py	works	Instantiates and runs cleanly.	no
ai-fashion-show-prototype.py	works	Instantiates and runs cleanly.	no
ai-mirror-of-truth.py	works	Covered by tests/test_mirror_of_truth.py (real emotion detection, Afterwords TTS, pluggable LLM, prosody — all headless-faked). Excluded from the generic smoke matrix (SKIP set) because it is a PrototypeBase subclass with a dedicated fixture-driven test.	no
apocalypse_experience.py	works	Instantiates and runs cleanly.	no
audio-reactive-fractal-forest.py	works	Instantiates and runs cleanly.	no
audio_mirror.py	works	Covered by tests/test_audio_mirror.py and tests/test_audio_mirror_fsm.py. Excluded from the generic smoke matrix (SKIP set) for the same reason as ai-mirror-of-truth.py.	no
augmented_audio_tours.py	degraded	Vision-based position detection is now explicit rather than silently swallowed: LLMBackend.generate_vision() exists as an optional capability (default NotImplementedError), and detect_position() logs a one-time warning and returns "unknown" when the configured backend lacks vision. Feature remains non-functional until a vision-capable backend is added.	no
chat-room-narrative-quilt.py	works (fixed)	Fixed in this sweep: corrected the self.l.lunar_tools_art_manager typo to self.lunar_tools_art_manager at construction. Constructs and runs cleanly headless.	no
collaborative-canvas.py	works	Instantiates and runs cleanly; .ip/.port read/write on ZMQPairEndpoint supported per Task 5.	no
collaborative_art.py	works	Instantiates and runs cleanly; .ip/.port writes supported by ZMQPairEndpoint (Task 5).	no
cosmic-soundscape.py	works (fixed)	Implemented in this sweep: InteractivePrototype subclass that transcribes a spoken phrase, maps keywords to celestial motifs/mood palette, generates a cosmic image via manager.image_gen, and renders it; graceful fallbacks when speech or generation is unavailable.	no
data-driven-cityscape.py	works (fixed)	Fixed in this sweep: missing api_keys.openweathermap no longer crashes construction (config.get instead of get_or_raise); _fetch_weather_data returns deterministic synthetic weather under LUNAR_HEADLESS, when the key is absent, or on HTTP failure — no network calls headless.	no
dynamic_visuals.py	works	Instantiates and runs cleanly.	no
emotional-landscape-generator-prototype.py	works	Instantiates and runs cleanly.	no
escape_room.py	works (fixed)	Fixed in this sweep: LLM intent parsing now guards gpt4 being None and generate() returning None, falling back to intent "unknown" with a logged warning instead of crashing.	no
evolving-cosmic-mural-prototype.py	works	Instantiates and runs cleanly.	no
generative-poetry-mosaic.py	works	Instantiates and runs cleanly.	no
interactive-storytelling-canvas-prototype.py	works (fixed)	Fixed in this sweep: replaced the from utils import record_and_transcribe_speech import (the top-level utils.py helper predates the tools layer and bypasses resolve()) in favor of self.speech2text.transcribe(duration=10), matching the convention used elsewhere (e.g. speech_activated_art.py); guarded the nonexistent manager.glif_api with getattr(..., None) (image visualization via Glif is out of scope for the current tools layer); fixed a self.lunar_tools_manager → self.lunar_tools_art_manager/self.glif_api typo in visualize_story().	no
interactive_storytelling.py	works	Instantiates and runs cleanly.	no
neural-transfer-music-visualizer.py	works (fixed)	Fixed in this sweep: SoundPlayer.stop_sound() added to the tools layer (with a FakeSoundPlayer no-op), and the prototype now uses non-blocking play_audio(...) instead of an invalid play_sound(..., loop=True) call.	no
real-time-glitch-art-lab.py	works	Instantiates and runs cleanly.	no
sentiment_analysis_display.py	works	Instantiates and runs cleanly.	no
speech_activated_art.py	works	Instantiates and runs cleanly.	no
temporal-art-gallery-prototype.py	works	Instantiates and runs cleanly.	no
time-shifted-echo-chamber.py	works (fixed)	The old xfail reason was a stale copy-paste from data-driven-cityscape.py — this prototype has no weather/network dependency. Constructs and runs cleanly headless against the fake recorder/player/keyboard tools.	no
virtual-cloud-chamber.py	works (fixed)	Fixed in this sweep: added FakeText2Speech to the headless tools layer and routed manager.text2speech through tools.resolve; the prototype now uses the correct generate(text) -> path contract, warns once and disables narration (visuals continue) if the LLM or TTS is unavailable.	no
virtual_time_travel.py	works (fixed)	The old xfail reason (.strip() on gpt4.generate()) was a copy-paste from escape_room.py and did not apply. Real run-loop fragilities fixed in this sweep: None text2speech and TTS InferenceError are now guarded/logged, and renderer.render is skipped when image generation returns None.	no
whispers.py	works	Instantiates and runs cleanly.	no

Summary

Totals across all 29 prototypes (27 legacy + audio_mirror.py + ai-mirror-of-truth.py):

• works: 28
• degraded: 1 (augmented_audio_tours.py — vision-based position detection explicitly degrades to "unknown" until a vision-capable LLM backend exists)
• needs-rework: 0

All former needs-rework prototypes were fixed in the July 2026 sweep; every xfail entry/marker has been removed and LUNAR_HEADLESS=1 pytest -q is fully green (199 passed, 0 xfailed).

Shared infrastructure: emotion classifier (Task 13, code half)

EmotionDetector (src/lunar_tools_art/emotion.py) now supports a real ONNX FER+ classifier (OpenCV DNN, 64x64 grayscale input, softmax over the 8 FER+ labels) alongside its existing Haar cascade face detector. It takes an optional model_path=None kwarg (default sourced from settings.toml [emotion] model_path); with no model file present, behavior is unchanged from the prior placeholder (has_classifier is False, confidence 0.0). scripts/fetch_models.py downloads and SHA-256-verifies the model weights; scripts/smoke_emotion.py is an on-machine webcam smoke test with an emotion-label overlay and FPS report. Not yet run on-machine: fetch_models.py (checksum is a documented placeholder pending a human running the real download once), smoke_emotion.py (webcam), and the flagship end-to-end Audio Mirror / Mirror of Truth runs with Afterwords — all deferred to a human per Task 13 Step 5.