nx_actor_role_audio_grader.nx

nx_actor_role_audio_grader.nx -- audio-scene-grading actor role. Per CARDINAL [[feedback-parallel-companion-multimodal-dnd-real-time]]: the parallel companion runs DIFFERENT modalities concurrently -- LLM token gen + audio grading + image grading all interleaved. This adapter wraps [[nx_audio_scene_quality]] (other-agent primitive) as a cooperative actor so the substrate can drive heterogeneous work, not just LLM clones. Cooperative phases: Phase 0 (INIT): no-op (reserved for future warmup) Phase 1 (GRADE): calls nx_audio_scene_grade with caller bins/rms/n_sources Phase 2 (EMIT): pushes AUDIO_FRAME message containing the verdict overall_score (Q14) as the payload handle Phase 3 (DONE): terminal Composes [[project-conductor-arc-phases-c-d-integrate-with-nishi-ai- 2026-05-19]] -- audio grading runs through the same actor + message + multiplexer substrate as LLM inference.