Events
A feature's events are typically what you are trying to optimize in a model driven feature. They provide the feedback that lets you know how well this particular version of a feature is helping or hurting the user.
Causal puts events in the same row as the context and outputs, making the format much more efficient and easy to use than a typical append-only data system. You can calculate the payoff for a data point using just the items in the row, even though corresponding events occur at different points in time.
Because Causal takes this unique approach to formatting data (all the data for an impression in a single row), it's much easier for a data scientist to create a training point.
A Note On Learning To Rank
For data nerds!
Causal uses memoization to group identical impressions together. That is, if you request a new feature impression, passing in the same arguments as a previous request in the session, you get the same result back.
This is important when you want to train a list-wise learn to rank model on data collected over several page impression. Using memoization, Causal can group all impressions showing the same content together, along with the complete set of items that received clicks. You should be able to create a listwise learn-to-rank data point off of each impression row in your feature table.