Most wireless standards were designed around a human receiver.

A person streaming audio can tolerate a 20 ms buffer. A person on a video call accepts occasional latency spikes as a normal part of the experience. A person downloading a file does not notice whether it arrives in 400 ms or 600 ms. Human perception has natural tolerance thresholds and wireless standards were engineered to stay within them.

This was a reasonable design assumption for the first three decades of wireless. It produced standards that work well for the applications they were built for: media streaming, file transfer, voice calls, web browsing, sensor telemetry.

A surgical assistant does not tolerate a dropped packet and retry. It is in the middle of a procedure.

An industrial cobot operating alongside a human worker does not have a graceful degradation mode for a 50 ms latency spike. A 50 ms spike at the wrong moment is a safety event.

An XR headset rendering a physical environment does not smooth over frame timing variation. Frame timing variation is nausea.

These systems share a common characteristic: they perceive the physical world, make a decision, and act in a loop that runs continuously, in real time, with hard timing requirements. The loop cannot be paused. The timing cannot slip. The action cannot wait for a retry.

This is a fundamentally different operating requirement from anything a human-tolerance wireless standard was designed to meet.

Most wireless standards are probabilistic. It is an architectural choice that makes sense for the applications they were designed for.

A probabilistic wireless system manages shared spectrum by negotiating access. Multiple devices contend for the same channel. Packets are sent; acknowledgements are expected; retransmissions are scheduled when packets are lost. Power is conserved by duty-cycling the radio while sleeping between transmissions, waking to check for traffic, sleeping again. Latency is managed statistically: the system guarantees acceptable average performance, not guaranteed per-packet timing.

For a human receiver, statistical performance is sufficient. Humans do not experience individual packet timing. They experience the aggregate; buffered, smoothed, rendered at human-perceptible timescales.

A machine that acts on sensor data cannot act on average latency. It acts on the latency of this packet, right now. A retransmission is not a minor inconvenience; it is a control loop that missed its timing window. A collision in shared spectrum is not a slight degradation. It is a gap in the sensor feed that the system was relying on to make its next decision.

The machine does not buffer. It does not smooth. It does not tolerate.

The natural response is to ask whether the problem can be solved by tuning. Lower the connection interval, increase the transmission power, add a dedicated channel, prioritise the traffic.

These are the right instincts. They are also insufficient.

The fundamental issue is not configuration. It is the mechanism by which the standard manages time.

Probabilistic wireless systems do not own time. They share it with every other device in the environment, with interference sources outside the system’s control, with the physics of a shared medium that no amount of protocol tuning can fully tame. The latency guarantee a probabilistic system can offer is always conditional: assuming no collision, assuming no interference, assuming the retry succeeds on the first attempt.