My understanding is that (all else being equal) experience from a prediction is linear in the size of your pool. Meaning that a full-pool prediction will teach the same as two half-pool predictions. This is a notable change from 2.0, where small-pool predictions had a learning bonus.
One consequence of this, though, is that waiting for full pool is slightly inefficient for learning. This is because the observation that brings your pool to full will probably be partially wasted, since your pool is at cap. If you were to predict more frequently, there would be no wasted observations, as it all transformed into experience. So if you're struggling with astro, I'd recommend not waiting until for a full pool before predicting. This will be slightly harder on any tool you're using, so don't do this with the ultimate tool you're trying to season.
