The assumption is that at some point it will get processed. The users are generally not expecting the flow to give up so it confuses them and is an extra condition for them to handle - which they mostly won't. This is something that showed up for a customer as their connectivity was so bad the retries were triggering even though there was no point.
Two things we should do:
1. Make it so there are infinite attempts. We should likely cap the time between attempts at whatever we have now for the last attempt.
2. If we can identify that the notary is not HA (i.e. not a clustered identity), there's no benefit to retries and so we should set the timeout to something very high as an approximation of infinity.