Newbie here so sorry if this is a stupid question, but why are samples trained independently of each other?
There are commonly pages with variation in structure on the same website, but I have to retrain the sample from scratch for each different page structure. Generally speaking under a common training profile I would be looking for the same fundamental data which may or may not be present on every page. If I wanted something different I could add an additional profile.
Seems like it would be easier to create a base training profile, run it and then train it to handle variations off the base profile.