Build some warnings into Amethyst that explicitly say when the relay misbehaves, even if Amethyst errors out because it cannot/should not recover.
You'll make Amethyst useful as an indicator for misbehavior with virtually zero overhead and there will be plenty of technically adept people who can read and understand the logs well enough to file a bug report with the relay. (And it shows a certain maturity and control on the side of the client.) 😁 

 Clatification: warning -> logging (possibly warning-level)