ShotSpotter's own public statements say that audio is retained for 30 hours, see e.g. https://www.soundthinking.com/faqs/shotspotter-privacy-faqs/. In the past they provided Albuquerque with a document that said 72 hours, but I believe they have been reducing that period over time. It may also depend on their specific contract with the law enforcement agency. Still, we know that they must retain recordings, because they do take requests from customers to reanalyze data that did not result in an alert.

For convenience, the most relevant text: "If the system misses a gunfire incident, police may contact the company to see if there is any audio or location evidence. In this case, only authorized ShotSpotter personnel with proper credentials can access sensor audio to search. Their search is limited to the 30- hour sensor storage timeframe."

ShotSpotter used to have significantly fewer privacy protections, and retention was indefinite early in the product's life. Fortunately, a combination of legal challenges and statutory privacy policies among their customers have lead to them significantly improving their privacy controls over time.

Is that, like provable though, thats my biggest issue. Cities are allowing a private company to mass deploy microphones all over (mostly lower income) areas and just _trusting_ them to not be keeping recordings.

> it is running on a very very short recording loop

I'm curious what your source for that information is

How would we know if this is true? For instance, is the source code subject to independent review, with verifiable builds made from that source so that all deployments have to show matching hashes? Or is more of a pinky-swear kind of a thing? If so, is the pinky swear even documented somewhere?