Human administrators are not a feature. Chargebacks and other forms of manual intervention allow a third party to diddle with the business of two transacting parties on the instruction of a single one of them and some idea/law/regulation about what is 'right'. To assume that this is a good idea is hopelessly juvenile and naive.

Try to UML-ify a system with manual intervention. Now modify it accepting the ideas that a) people are imperfect; and b) people work on their own incentives which may not (probably won't) be aligned with the users of the system. The manual intervention bit is now a black hole of undefined behaviour with nothing but a wing and a prayer to say it might actually do something useful.

Try and get that past QA.