LiveKit vs mediasoup

The Rust line spent most of this window on single-PR point releases, nearly all of them subchannel plumbing for DataConsumers over pipe transports plus the occasional crash fix. That pattern broke with 0.26.0, a seven-PR reworking of how RTCP Sender Reports are generated, moving them off RTP packet arrival time and onto an estimate of when media was captured. 0.27.0 continues in the media path with a breaking change limiting simulcast and SVC to the preferred temporal layer. The 0.24.0 release earlier in the window also carried real security work — constant-time MAC comparison plus fixes for an OOB write and an SCTP integer overflow.

The subchannel thread that dominated July has gone quiet and the media path has taken over. First timing correctness — RemoteClockOffsetEstimator and RemoteCaptureTimeEstimator gave the worker a notion of remote capture instants it did not previously have — and now layer selection, with 0.27.0 constraining which temporal layer simulcast and SVC will actually deliver. Both are behavior changes in what an SFU sends downstream rather than additions to its API surface, and 0.27.0 is labelled breaking, so integrators are being asked to absorb them.

With two consecutive releases changing media-path behavior, the near-term releases are likely to be corrections shaking out the new timing and layer-selection paths across simulcast and pipe transports. The capture-time estimators remain unused beyond Sender Reports, which leaves them the obvious foundation for further synchronization work.