Retainer xp has been constrained in a few ways. First, while retainers can transfer xp between each other, this has a cap of only matching xp that was directly transferred in by a player. In essence, they can't transfer xp gained by training, only xp directly donated.

I initially toyed with a system of having a training cap for retainers, but I didn't like how that devalued higher levels of animal ken/teaching. So what I'm doing instead is having the difficulty for training increase based on their xp on hand (only unspent, not total they've earned lifetime) and their permanent quality level. You can now spend AP on retainer training to lower the difficulty, and their training difficulty remains lowered until a successful roll is made. The lowest you can bring the difficulty to on any roll is -10. Current difficulty is displayed on @retainer/viewstats.

The formula I settled on is (unspent_xp_penalty + agent_level_penalty) * difficulty_multiplier - conditioning. The unspent_xp_penalty is half their xp minus their agent quality level times 15, with a minimum of 0. The agent_level_penalty is their agent quality level times 5. The difficulty multiplier is 0.75 for an animal retainer, 1.0 for non-animal. Conditioning is an arbitrarily large integer based on the sum of all AP spent so far to lower difficulty without success. After adding all those together, the difficulty is lowered to at most -10. So, for example, Steve the Guardsman has 180 unspent xp and is level 4, with 35 AP spent on previous rolls. That would be (((90 - 60) + (20))* 1) - 35, for a difficulty of 15. If Steve was turned into a rabbit, that would be (50 * .75) - 35, for a difficulty of 2.

The idea is to incentivize two behaviors: spending xp and resources to increase a retainer's stats/skills rather than just letting its xp grow indefinitely by doing a command run every week, and to provide another incentive to spend AP and/or seek out the most capable trainer for more difficult rolls.