nx_room_simulcast.nx

nx_room_simulcast.nx -- X-ROOM (GEN-ROOM-simulcast-quality-layers): sovereign DETERMINISTIC INTEGER simulcast layer allocator -- the SENDER-side companion to nx_room_abr (the receiver-side ABR controller). ABR decides which quality layer each receiver PULLS; simulcast decides which layers the sender PRODUCES, given a finite uplink budget and the set of layers receivers actually demand. EXCEED axis vs Zoom/LiveKit/WebRTC simulcast (honest): two deterministic, audit- replayable levers a heuristic encoder leaves on the table -- (1) DEMAND-DRIVEN PRUNE: encode ONLY layers some receiver requests; a layer nobody pulls is never encoded = uplink saved (naive simulcast always pushes all spatial layers). (2) BUDGET-BOUNDED SHED: if the demanded layers exceed the uplink budget, shed the highest-bitrate layers FIRST, deterministically, until the encoded set fits -- never over-subscribe the uplink (which a budget-unaware encoder does, causing the very congestion ABR then has to fight). MODEL (bitrates kbps): ladder[] = simulcast layers low->high (DATA, rule 25). demand mask = bit i set iff some receiver requests layer i. allocate: start from the demand mask; while sum(encoded bitrates) > budget, clear the highest set bit (shed) and count it. out[0]=encoded mask, out[1]=encoded kbps, out[2]=shed count. naive baseline = always all layers = sum(ladder). main() is the SELF-VALIDATING GATE. Evidence -> knowledge/status/room_simulcast.log. HONEST SCOPE: this is the produce-set ALLOCATION (prune+shed); per-receiver fallback re-mapping after a shed is a named deeper rung, not claimed here. license_tier: ORIGINAL