2. We are also interested in a question that seems to have been asked here quite a bit before which is about audit. It seems like the current best practice is to add any context data required into the context of the tuple, and use the watch api to store this somewhere else for query. Is this true? and what limits are there on the size and content of the caveat context?