Review for NeurIPS paper: Group-Fair Online Allocation in Continuous Time

Weaknesses: (1) The first weakness of this paper is that the online learning algorithm makes an assumption about full-feedback after a certain time-delay. This seems unreasonable to me in the applications, since it is a-priori unclear how you would get the quality estimates for other groups given a task. The paper starts with a pretty general model where the tasks across different time-steps can be pretty heterogenous. Is this assumption so that the technical analysis goes through? In that case, please mention that. Also discuss why it may be okay if the tasks are similar.