The fastest way to make a presence product dishonest is to start naming who is in the room. We get asked about guests, pets, and multiple people more than any other technical question, so here is the full answer, including the parts that cost us sales.

What the sensor actually measures

A WiFi field in a room changes when something moves through it. That is the entire signal. It does not carry identity, count, or intent. A cat crossing the sofa, a visitor leaving, and the person who lives there walking to the kitchen all perturb the field in ways that are, at our sampling resolution, the same class of event. We report "the room changed". We do not report "two people" or "she has a visitor".

Why we did not ship a pet filter

Classifying a pet from coarse signal disturbance would require a model trained on labelled data from real homes, and it would be confidently wrong often enough to matter. A filter that suppresses some cat movements and occasionally suppresses a person is worse than no filter, because the family would reasonably assume the display is now people-only. So we publish the limit instead: if a pet moving around the room makes the display useless for you, this is not the right device. That sentence belongs on the product page, not in a support ticket six weeks later.

Guests, and the quiet-room problem

When guests come over, the room changes more often. The board shows more changes. It does not know why. The same honesty applies in the other direction: a person reading a book for two hours looks very close to an empty flat on a coarse signal, so we give "still but present" its own state rather than collapsing it into "no one". We would rather show an ambiguous state than a clean one we cannot justify.

Comparison with a door sensor

A door contact knows a door opened. It is cheaper, simpler, and completely blind to someone who spent six hours in the living room without opening anything. That gap is why we used the WiFi field. But the comparison cuts both ways: a door sensor makes a narrow, well-defined claim, and any attempt to turn our broader signal into a story about a person's day would be an overclaim. We keep the claim narrow even though the signal is broad.

What we refuse to add

No people counting, no identity, no routine scoring, no "unusual day" flag, no push notification. Each of those requires the product to assert a reason for a change, and the signal does not contain reasons. We also do not send alerts, measure heart rate or breathing, or claim to detect falls. This is not a medical device, it does not diagnose anything, and it provides no emergency response — in an emergency, call 120. It is not a replacement for a caregiver or for the daily phone call.

The interface that follows

Four states: moving, still but present, no one, no signal. Two numbers: how long since the room last changed, and whether the link is alive. When power is cut, the router reboots, or the link drops, we show a separate state instead of leaving the last good reading on screen. The unit samples about once a second from a real occupied apartment; the public board is at annest.cn/live/ and reports live, not from a recorded loop.

The unit is 499 CNY including the first month of service, then 30 CNY per month. No camera, no microphone, no wearable on the person being observed. We wrote this list of limits before we wrote the landing page, and we would rather lose a sale than explain a confident guess that arrived too late.