Explicit render-scene workflow

OpenJOC can render a portable, static scene made from caller-bound mono WAV sources and a locally supplied SimpleFreeFieldHRIR SOFA file. The scene is versioned as openjoc.render-scene.v1; it does not contain JOC objects, ReconstructionBasis rows, OAMD slots, or backend settings.

{
  "schema": "openjoc.render-scene.v1",
  "sample_rate_hz": 48000,
  "source_semantics": "explicit_spatial_sources",
  "sources": [
    {"id":"voice","input_wav":"audio/voice.wav","start_sample":0,
     "position":{"x":0.0,"y":1.0,"z":0.0},"gain":1.0}
  ]
}

Source paths are relative to the scene file. Absolute paths, parent traversal, symlink escapes, duplicate IDs, unknown directions, and source-WAV sample-rate mismatches are rejected before output promotion. Supported source WAVs are mono PCM16/24/32 and mono IEEE-float32; they must match the scene rate. OpenJOC does not resample, normalize, clip, or dither source audio.

Inspect a supported SOFA file first:

openjoc sofa inspect listener.sofa --json

Render with an explicit backend:

openjoc render-scene scene.json --binaural-sofa listener.sofa \
  --backend direct --output render-direct
openjoc render-scene scene.json --binaural-sofa listener.sofa \
  --backend partitioned --partition-size 256 --output render-partitioned

The output directory is transactional and contains binaural.wav (stereo IEEE-float32, FL then FR) and render.json (openjoc.render-result.v1). Output length is the scene input timeline plus the complete causal HRIR tail (N + M - 1); no leading latency or tail trim is hidden. Backend selection is explicit and never automatic.

The SOFA boundary is intentionally narrow: SimpleFreeFieldHRIR, admitted versions 1.0/1.1/1.2, and the read-only NetCDF classic CDF-1 or NetCDF-4/HDF5 subset. HRIRs are converted to the scene sample rate; inputs already at that rate retain their source coefficients bit-for-bit. Other SOFA conventions, interpolation, nearest-direction fallback, moving sources, and downloads are not supported. Users remain responsible for the licensing and provenance of locally supplied SOFA data.

When rates differ, conversion preserves convolution gain and includes a common causal filter delay in both HRIRs and the output tail. The result manifest adds this delay to the backend's algorithmic_latency_samples. See the conversion delay and HDF5 chunk limits in Binaural and SOFA.

The workflow is independent of unresolved JOC semantic binding: joc_semantic_binding is unresolved_not_used in the result manifest.