Known limitations and release boundaries

• Execution supervision is macOS-specific; Apple Silicon/M4 Max is the measured environment. Linux, Windows and every Intel Mac are not certified targets.
• FFmpeg/FFprobe are external dependencies. A listed codec or hardware feature is not a successful encode or proof of GPU use on your machine.
• Native build, patch, engine activation, immutable release and trusted exec commands need the full source checkout and fresh local acceptance evidence. The redacted historical indexes cannot activate an engine.
• No prebuilt native engine, benchmark media, private PGO profile or runtime database is distributed here. Research scripts require their own retained corpus and receipts. Path configuration alone does not recreate that lab.
• Phase 11's complete HD workflow passed its bounded timing gate. Direct HD/raw sample requirements failed. The candidate is experimental, with no complete new-prefix FATE/soak/install promotion acceptance. Original local 0.6.0 remains the verified lab default; public 0.6.1 is a source/portability release.
• CPU instruction reductions, metadata byte differences and warmup timing are not compression or general speedup claims. Different experiment epochs are not pooled.
• HDR, interlaced, variable-frame-rate, subtitles, audio and container policies have explicit supported bounds. Unsupported combinations fail closed. A successful tone-map is not a perceptual HDR mastering certification.
• Output validation is bounded by the selected workload and local decoder. It does not establish universal perceptual quality for every camera or codec.
• Resource limits are sampled watchdogs. Abrupt power/storage loss and kernel/system-wide scheduling behavior remain outside the application's control.
• Some diagnostics and historical reports are Bengali. English setup/reference guides and stable error identifiers are available.

See historical gap register, readiness, benchmark method and report archive.