So here's the thing.
The intent of the custom onSuccess listener on your page is to stay on the same page without refreshing and show a Thank You message. Ergo, there should never be a pageview with the ?aliId query parameter in the URL, because that only happens with the default (non-customized) onSuccess behavior (or if you deliberately emulate that same behavior in custom JS).
However, the onSuccess listener is added as part of a larger block of only semi-related code (starting on line 399) rather than standing alone. If any fatal error should happen in that code before the onSuccess is added (it's only added at the very end) then the rest of the code will never execute. There are enough different functions in there that certainly something could go wrong under certain browser conditions. If that happens, the form will still submit, but it'll be using the default onSuccess behavior -- which is to refresh the IFRAME with the ?aliId and show the form again. That could well lead someone to think the form didn't do anything, as I mentioned.
As for what in those 100 lines of code could sometimes, but not always, error out, it's hard to say. I see some very minor oversights (looking for fields starting with "utm" fields in the whole document, instead of just the current form, for example, and not securing postMessage as much as I'd like) but nothing that calls out as fragile. But I can all but guarantee that's what's happening: something else is erroring out, so the listener doesn't get overridden, so the end user thinks the form didn't get posted and tries again.