Glass-to-glass measures the entire chain end-to-end: lens, sensor, ISP, capture, encoder, contribution protocol, origin, packager, CDN, last-mile network, player buffer, decoder, compositor, and finally the panel. Each stage adds a budget. For HLS with 6-second segments and a 3-segment hold-back, that budget is 18–30 seconds and most of it sits in the player buffer. For LL-HLS with 1-second partial segments, you can cut it to 2–5 seconds. WebRTC and Media over QUIC reach 200 ms – 1 s.
Operators usually measure glass-to-glass with a stopwatch trick: point a camera at a millisecond-accurate clock display and let the player render that camera's stream on a second monitor next to the original clock. Photograph both clocks with a third camera and read the difference. There are also dedicated SDI/IP probes (Bridge Technologies, Telestream, EZ Drummer) that calculate it from timecode metadata.
The right number depends on use case. Cinema-grade broadcast tolerates 30 s and gains compression efficiency. Sports betting requires under 2 s or it is unusable. Auctions, second-screen quiz games and any "react together" UX need sub-1 s. Glass-to-glass is the single number to negotiate with stakeholders before architecting; everything else follows from it.
