Write better AI detection triggers
A useful trigger describes something the image or audio can actually show. Give Snitch one observable condition, then test both a matching scene and a normal scene before starting a longer session.
Name the subject and the state
“A cat is on the counter” identifies an object and a location. “Something bad is happening” supplies neither. For delivery monitoring, choose “a package is on the doormat,” not “my order is here.” For audio, describe the sound rather than a person’s intent.
Match the condition to the selected area
If the condition mentions a door and a package, both need to be visible in the crop. A tight box is useful only if it still includes enough context. Read your condition while looking at the chosen region: could someone answer it from that view alone?
Change one thing at a time
Try a matching event, an empty scene, and a plausible distraction. Review the explanation, confidence, and replay. Change the wording, framing, or sensitivity separately so you can tell which change helped. A confidence value is part of an AI result, not a guarantee of correctness.
Retest when the scene changes
Lighting, moved furniture, new background noise, and camera position all affect results. Keep the phone awake and online with Snitch open. Use everyday, low-stakes observations and expect occasional missed or incorrect matches.