Every live stream sits somewhere on a trade between how far behind reality it runs and how reliably it plays. The two cannot be improved together, because the same mechanism produces both.
A buffer is stored time
Players do not display video the instant it arrives. They accumulate a few seconds first, then play from that store while more data continues to arrive behind it.
That stored time is what allows playback to continue through a network interruption. As long as the gap is shorter than the buffer, the viewer sees nothing.
Reducing latency means holding less video in reserve, which directly shortens the interruption the player can absorb before the picture stops.
Networks vary constantly rather than smoothly
Delivery over ordinary connections is uneven by nature, with packets arriving in bursts and occasional gaps caused by congestion anywhere along the route.
Average bandwidth being sufficient does not mean the stream arrives evenly, and a player with a small buffer is exposed to every one of those short interruptions.
This is why a viewer with a fast connection can still experience stalling on a low-latency stream while a slower connection plays a higher-latency one perfectly.
Encoding introduces its own delay
Compression works by describing frames in relation to their neighbours, which requires the encoder to hold several frames before it can produce output.
Configurations that look further ahead compress more efficiently and add more delay, so bandwidth efficiency and latency are also in tension.
Very low latency setups restrict how far the encoder can look, which raises the bitrate needed for the same quality and shifts the cost onto delivery.
Different formats want different settings
Interactive streams where a host responds to chat need low delay, because a conversation with a long gap stops functioning as a conversation.
Broadcast-style streams with no live interaction gain nothing from low latency and everything from reliability, so they run with generous buffers deliberately.
Choosing the wrong configuration for the format is a common fault, producing streams that stall constantly in exchange for responsiveness nobody is using.
Adaptive delivery moves the trade rather than removing it
Streams are commonly published at several qualities, and the player drops to a lower one when the buffer starts to shrink instead of stopping entirely.
That converts an interruption into a visible quality reduction, which most viewers tolerate far better than a stall, but the underlying constraint has not changed.
At very low latency there is less time to detect trouble and switch, which is why adaptive systems are less effective exactly where they would be most useful.