@ Syth:
Currently the miner tries to re-connect too often when it things that connection is lost. Next version will reduce the number of attempts and also give the connection more time for recovering.
From my experiments with my 24 hours run the earnings were accurate (on Optiminer pool) according to the pool luck. On minexpool I did not test but I know that difficulty is higher there, so on the long term the earnings should fit, but it can be more volatile on the short term. As always it is important that the average sol/s detected by pool (over one hour or more) fit to whan the miner displays on command line. Then the miner works well. Everything else is in the responsibility of the pool.
Adding something like --retries and --retry-delay would be nice or even backup pool options.
If you just allowed the program to exit after failing to reconnect the specified amount of --retries consecutively, it would allow us to simply run a new instance trying another pool automatically.