The senior surveillance lead notices what the alerting system doesn’t know how to name. A pause running slightly long. A route through the floor that’s technically ordinary and completely wrong. Two innocent actions made meaningful by their order. That judgment took thousands of quiet hours to build, and when she retires it walks out the door, which is why institutions ask me about capturing it.
I’ll do that work, but only under five clauses, and I put them in writing.
First. No autonomous adverse action, ever. The system retrieves related events, organizes incident notes, drafts chronologies, and points a reviewer at inconsistencies. It doesn’t accuse. A fluent narrative can weld a weak association into a story, and a number can borrow authority just by being a number, so outputs stay evidence for a person rather than verdicts from a machine.
Second. No hidden scoring of employees. Knowledge capture that mutates into covert productivity ranking poisons the well it’s drawing from, and the gaming commission, labor obligations, and the institution’s own investigative rules govern everything the system touches.
Third. No unexplained match. An identity or pattern assertion that can’t show its sources is gossip with a database, and it doesn’t ship.
Fourth. Capture happens with the expert, never to her. She decides which signals matter, where the tools mislead, what she verifies before believing, and what she refuses to infer. The teachable parts, the questions asked, the traps known, the camera limitations, the reasons a plausible pattern got dismissed, become reviewed procedures and retrieval paths. The untranslatable parts stay hers, and pretending otherwise produces confident nonsense.
Fifth. Test on quiet footage, not just incidents. A system tuned exclusively on dramatic outcomes learns that every story ends badly, and ordinary nights aren’t drama; they’re most of the evidence base.
The goal was never a machine that sees like the veteran. It’s the next shift finding the relevant record faster, examining it with more discipline, and inheriting more of her method than a binder ever held. Institutional memory deserves an architecture, and the person who built that memory deserves a say in it.