Number of confirmations going down?

[jon_smark]

I've witnessed something puzzling on the Testnet: three blocks I mined had their number of confirmations suddenly go down.  One of them jumped from 285 to 263, though it has since monotonically increased again, as one would expect.  I thought that the number of confirmations could only increase, unless the block finds itself in an orphaned fork, in which case the number of confirmations becomes meaningless. So, what may explain this sort of situation?

[TangibleCryptography]

The number of confirmations is the confirmations of the current longest chain.  
Orphaned blocks are completely ignored so a tx which is included in a block in only one chain will revert to 0-confirms in the event an alternate chain becomes the longest.  It is possible for a tx to be included in blocks of different heights on more than one fork.  

If your tx was included in block# C-285 (C being current block height of the longest chain) in one fork and C-263 in another fork when the second fork replaced the first one as the longest chain your node would recompute the number of confirms and show 263.

Note: such a deep replacement in the main-net is highly unlikely outside of a 51% attack or a subnet becoming islanded for an extended period of time (network breakdown).

[SgtSpike]

I'd be interested in knowing why such a fork took place.  Is someone testing a 51% attack method?