This kind of thing has happened to me before. My surveys make use of two webhooks: one to P.A. and another to an external web application. I sometimes find that one webhook will fire while the other does not; that is, I can see a few instances where a payload has arrived in the web application without a corresponding P.A. flow trigger, and vice-versa. I have heard from some engineers at Esri that this kind of buggy behavior in their webhooks is known, but perhaps not well-understood. It doesn't happen often, but it does happen. That was all to commiserate with you! Now to try and answer your questions:
1. I am unaware of a way to see any kind of log of webhook activity, and I have spent a little time searching for them but not much as the times this has happened to me were pretty inconsequential. I wouldn't be surprised if they do exist! If they do, there are almost certainly ways in the backend to get them. I think reaching out to Tech Support would probably be your best bet for this. Also, I would check your flow trigger parameters in both P.A. and on the specific survey's Settings > Webhooks page, and make sure that you are having the webhook fire when any event happens in the feature service, and are including all details of the event.
2. I personally have found that P.A. is extremely finnicky when it comes to keeping connections to Survey123 active. No idea why that is the case, but basically every time I want to make a new flow, or even when some survey hasn't been utilized in a couple months (e.g. when the field season ends), I have to create a whole new connection between P.A. and Survey123 in order to restore the flow. To start, I would definitely check the trigger of your flow and make sure that the Survey123 connection is actually still active. If you have the time, you could do some stress-testing with your survey and watching P.A. to see whether or not each submission is caught - if you can do this, start with the sparsest-possible payload (i.e. fill out nothing in the survey) and see if that registers in P.A. Then, scale up your testing by creating incrementally larger payloads when you submit, up until you feel like your test submissions are roughly accurate in size and complexity with a live submission. You may find there is a tipping point. As one example, I know that P.A. doesn't deal in payloads that have more than 1,024 "nodes" in the schema - this one has been a gotcha for me.
Good luck! Feel free to DM me if you want, I may be able to help more. I have done more screwing around with P.A. and Survey123 than anyone should have to in life.
