However when a Place is first automatically assigned to a Visit, it will be done just before the recording engine goes into sleep mode. (It needs to know what Place you’re at to know how likely you are to leave, thus how deeply it should sleep). So when it first assigns a Place, the Visit will likely only be a few minutes long.
However when you open the app later, the automatic Place assignments will be reassessed, and if the previous best match is no longer appropriate, it’ll get automatically changed. (Note that this doesn’t happen for manually confirmed Place assignments - those stay as you confirmed them).
So if you open the app to check the Place assignment shortly after arriving somewhere, it might still have an inappropriate assignment, if there’s another commonly visited Place nearby that better matches short visits. But when you reopen the app to check the timeline later on, it will be automatically updated to the better match.
More specifically to this problem, it’s likely that some Visits have been incorrectly confirmed as being to the park, which will be confusing the models. (The models base their choices only on assignments you have explicitly confirmed - automatic assignments don’t influence the models at all).
The easiest way to fix this is to go into the details view for one of those park visits, scroll down to “Common duration”, and tap through to that view. The bar graph in there lets you tap on different common durations, then below you can tap “Visits” to see a list of all visits that fall within that duration. Then you can quickly identify visits that have clearly been incorrectly assigned, and correct them. For example if you’re only ever at the park for 5-20 minutes, but there’s a bar for 6 hour visits, you can know that all those 6 hour visits are incorrectly assigned, and now have quick access to correcting them.
Once those corrections are done, the models will be much less likely to make the same mistake again. (Though it will take up to a day for the models to get updated in the background).