# Research Archive This directory preserves implementation research and hard-won calibration lessons. It is allowed to mention external tools, papers, vendor docs, and failed paths because forgetting those days would be an expensive little act of self-sabotage. Research is not runtime authority. Use it to shape native implementation decisions: - direct camera and audio drivers; - pinned/native buffers; - GPU feature extraction, flow, matching, depth, and fusion; - Fensalir D3D12 compute/render paths; - Faust/native DSP; - OBS-facing program outputs. Visual/GPU research entry points: - [[d3d12-stereo-depth-provenance-2026-05-29|D3D12 Stereo Depth Provenance]]: libSGM is the current permissive north-star reference for a future Fensalir-owned D3D12 compute stereo-disparity lane. External CUDA/TensorRT projects are provenance only, not runtime authority. Audio timing research entry points: - [[chirplet-sync-decoder/summary|Chirplet Sync Decoder Summary]]: chirplet transform notes, why dense matched-filter search was cut from the hybrid hot path, and the current dechirp plus FFT/Goertzel decoder shape. - [[live-adaptive-sync/summary|Live Adaptive Sync Summary]]: passive program-audio synchronization and sample-rate-offset estimation. - [[feedback-calibration/summary|Feedback Calibration Summary]]: online room/speaker/mic response calibration. - [[ambisonic-aquasynth/summary|Ambisonic AquaSynth Summary]]: spatial audio and Ambisonic reference material; useful background, not a license to call unsynchronized mics a coherent field. Do not promote mirrored examples, old probes, or external library demos into the Mimir hot path just because they exist here. If research points at a concept, rebuild the live version in the Mimir/Fensalir/native stack with a clear owner and invariant.