Model Response: Camera A sits on the vault’s only point of entry, so it records both ingress and egress. An adversary is captured at the moment of entry — badging in on a credential that is not theirs, or slipping through behind an authorized employee — and captured again on the way out, so the footage shows what was carried out of the vault and by whom. That is exactly the record an after-incident investigation needs: which door, which person, what left the room. Camera B does the same job for the exterior: it records the roll-up door being forced or quietly propped open for a later entry, with the adversary’s face and hands in frame at the moment of entry. Why the parking lot fails: a wide view captures presence, not action. It establishes that somebody was on site, but not which door they used, whether they badged at all, or what they were carrying when they left — and at that distance a face is not identifiable, so it does not even establish who. B.1 asks you to weigh visual coverage, angle and tamper resistance by asking what a camera in that spot would capture an adversary doing; a camera earns its position by that test, not by how much ground it covers.

Model Response: Claim 1 fails on placement. A motion sensor detects only where traffic is unexpected. Reception, the main corridor and the stairwells are high-traffic space during the day, so a sensor there fires on legitimate movement all day long; the resulting stream of false alarms teaches whoever receives them to ignore the alerts, and a sensor whose alerts are ignored has stopped functioning as detection even though the hardware works perfectly. The sensors belong where few authorized people ever go: the locked cabinet alcove holding the engineering file server, the vault interior, and the below-grade optics bay, which today has no detection of any kind. Claim 2 fails on the remedy, which is a different error. Alarm frequency is not evidence of effectiveness — it is usually evidence of bad placement — and dropping the sensitivity across every sensor does not remove the false alarms so much as it buries the real intrusion along with them; you cannot desensitize your way to a working detection control. The fix is placement plus scheduling: put the corridor and stairwell sensors on a schedule so they are live only outside business hours, when any movement they see really is unexpected, and leave the server alcove, the vault and the optics bay live around the clock. Claim 1 is an error about where; claim 2 is an error about how.

Model Response: 3a. A patrolling officer buys time pressure. The officer is harder to plan around than a fixed post because the route is unpredictable, so an adversary cannot know how long a break-in has before somebody walks around the corner, and that uncertainty is a deterrent in its own right. What the patrol gives up is continuous coverage: unlike a stationary post, every point on the route sits unwatched between passes, and a fast attack on the loading dock can start and finish inside a single interval. The design has to answer that directly — randomize both the route and the interval so the gap cannot be timed, and cover the gap in time with a control that watches while the officer is on the far side of the building, such as a door contact on the optics bay entrance, the one funded area with no detection on it at all.

3b. Curtis is using the wrong kind of evidence. Four months with nothing reported is not proof that nothing was attempted; it is a count of what Delmar was able to detect, and Delmar’s detection coverage is incomplete by design — the exterior and the below-grade bay have no detection funded on them, so an intrusion there would leave no record for anyone to report. Absence of evidence is not evidence of absence when you already know the sensor set has holes in it. It also confuses two different functions: five of the six installed controls are preventive, and prevention and detection do different work — a lock, an enclosure or a flood barrier can stop an attack while producing no record that one ever happened. Cutting the officer and the feeds would remove the only part of the plan that generates a record at all, which is precisely what an after-incident investigation runs on.