A Jellyfin transcode is generated as it is sent — no Content-Length, Range
ignored — so every interruption restarts it from byte zero, and three
concurrent downloads are three ffmpeg jobs on the server. Measured on the
tablet: a direct copy moves 2.06 GB in 142 s with no retries, a transcode
crawls at ~1 MB/s and cannot resume. The transcode is only ever requested
because of the audio track.
So: always fetch Static=true, and re-encode the audio on the device when the
source carries something the renderers cannot decode.
DR-171 keeps its diagnosis — a downloaded film played as picture in silence,
and offline there is no other source to fall back to — and loses its remedy,
which explicitly accepted the loss of byte-range resumability. Its other
finding survives and constrains this spec: a downloaded file outlives whatever
experimentalNativeVideo was set to when it arrived, which is why the fix is to
the bytes on disk rather than to one renderer.
Records the rejected alternatives with the specific reason each fails, and the
decision to accept HEVC staying HEVC (Android-first; recoverable by a setting,
unlike silent audio).
