Baseline refresh: when a flagged change becomes the new normal
How to decide when a confirmed change at a site stops being an anomaly and gets folded into the pattern-of-life baseline.

A baseline is the accumulated record of what a site does, day over day: when trucks show up, where the vehicles park, which bays sit empty on weekends, how long the dwell times run. Build that from repeat passes and a new collect either matches the pattern or it doesn't. When it doesn't, you get a flag.
The part that trips up a lot of monitoring setups is what happens after the flag fires. A flag tells you a specific pass didn't match the baseline, say, three extra vehicles in the yard, or a bay that's normally empty showing activity. It doesn't tell you whether that's a one-off or the start of a pattern you haven't accounted for yet.
What counts as "normal" in the first place
Normal is a statistical statement, not a visual one. If a lot bay sits full five days a week and empty on the sixth, that pattern is the baseline working as intended, not a day-six anomaly. The trouble starts when a site changes its behavior for reasons that have nothing to do with whatever you're watching for: a new shift schedule, a contractor rotation, a seasonal slowdown. The imagery looks different, the flag fires, and the underlying cause is mundane.
That's fine the first time. It's a problem if the same "anomaly" keeps firing on every pass for three weeks straight. At that point your baseline is describing a site that no longer exists, and every new collect is getting compared against a stale picture.
The fold-in decision
Folding a flagged change back into the baseline means you stop treating it as a deviation and start treating it as the new default pattern. That's a judgment call, and it should stay one. A few things worth checking before you make it:
Has the change held across enough independent passes to rule out noise, like a parked convoy or a one-day event, rather than a shift in routine? One or two matching collects isn't enough. A run of consistent passes over a couple of weeks is a better signal that the site has settled into something different.
Is there a known cause? Confirmed construction, a change in unit rotation, a facility expansion visible in the imagery itself. A fold-in tied to a confirmed cause is defensible later if someone asks why the baseline moved. A fold-in based on "it's been quiet for a while" is harder to stand behind.
Does the change match the shape of what you were actually watching for? This is the one people skip. If the thing you're monitoring for is exactly the kind of activity that just got folded into "normal," the fold-in created a blind spot. Calling it housekeeping doesn't change that.
If those checks hold up, update the baseline and move the anomaly history forward. If they don't, leave the flag open and keep watching. A baseline that updates too eagerly is almost as useless as one that never updates at all, because either way you stop getting a reliable signal when something genuinely new happens.
Baseline drift and why it's worth tracking on purpose
Drift happens whether or not anyone decides to fold a change in. Sites change slowly: vehicles get replaced, a new access road goes in, and a baseline that was accurate in March can be wrong by October without any single dramatic event marking the shift. Updating the baseline should be a deliberate step with a record attached, not something that happens by accretion because nobody went back to recheck an old flag.
That record matters more than it sounds like it should. Six months from now, when someone asks why the current baseline treats a certain vehicle count as routine, "we confirmed it against four weeks of passes in June" is an answer. "It just stopped getting flagged" is not.
This is the part of pattern-of-life work that doesn't show up in a single image review, which is exactly why Activity-Based Intel is built around daily collects stacked into a running baseline rather than a snapshot. If you're deciding whether a flagged change at your site has earned its way into the new normal, a daily-baseline setup built for that exact decision is worth a look.