Once we have batch proofs, the budgets for "internal" vs. "user-provided" batches would be different. Specifically, because we don't need to generate batch proofs ourselves, we could be much less conservative on the number of txs that people include in their batches. We will need to check things like the number of output notes produced, number of accounts updated etc. but if people want to batch 1K transactions into their batches while staying within these limits, we don't really care too much.
Its unclear when a good time to do this might be; with training wheels enabled we have to re-execute and re-prove these so its not zero cost.
Its unclear when a good time to do this might be; with training wheels enabled we have to re-execute and re-prove these so its not zero cost.