Limitations
What this system cannot do
This page exists because the honest version of this pitch is more useful to you than the confident version, and because a school that misunderstands what it bought will build a response protocol on top of that misunderstanding.
It cannot detect a concealed weapon.
The detector works on what a camera can see. A firearm in a waistband, a bag, or under a coat is invisible to it, and no amount of model improvement changes that. Anyone describing a camera system as detecting concealed weapons is describing something else.
Detection is not prevention.
At best this shortens the gap between a firearm becoming visible and the right people knowing about it. It does not stop anyone. What you do with those seconds is a protocol question, and the protocol matters at least as much as the detector.
It produces false alerts.
Phones, umbrellas, power tools, caulk guns, camera straps, toy and prop guns, and plain shadows all get flagged sometimes. Documented incidents in this industry include armed police responding to a bag of crisps. Your site's own measured rate is on your dashboard permanently and next to every response setting.
It misses things.
Distance, low light, odd carry angles, partial occlusion, and a camera pointed the wrong way all reduce recall. A camera that cannot resolve a firearm at its working distance is rated unsuitable during setup and excluded rather than quietly contributing misses.
It depends on your network and your roster.
If the appliance loses connectivity the site is not being watched, and you will be told so. If the people on your roster left in June, alerts reach nobody. Both are failure modes with their own alarms in this product, because both are more common in practice than a model error.
We do not publish an accuracy number.
A single percentage is meaningless without the operating point, the distance band, the lighting, and the camera it was measured on. What we publish instead is what real deployments measure, below, and what your own site measures, on your dashboard.
What we do claim
That when a firearm is visible on a camera we analyse, this system is likely to notice it quickly and put the frame in front of the people you chose — with the ambiguity labelled rather than hidden, a way to stop the response if it was wrong, and a record of everything that happened afterwards.