Problems in Dukascopy data

I downloaded the H1 data from Dukascopy. The data seems accurate, but there are
some strange candles that I do not understand.

I will start with the most recent problem: there is a "Sunday" candle for 2014-10-26.

26.10.2014 19:00:00.000,1.26688,1.26688,1.26688,1.26688,0.0000

26.10.2014 20:00:00.000,1.26754,1.26775,1.26728,1.26738,2233.0901

26.10.2014 21:00:00.000,1.26738,1.26793,1.26719,1.26757,2361.6299

The timestamp is 26.10.2014 20:00:00.000. In NY time it is 2014-10-26 16:00:00, in
"forex time" it is 2014-10-26 23:00:00. 2014-10-26 was a Sunday, so no candle should
have been recorded for that day. There are other similar candles in other years:

31.10.2010 20:00:00.000

30.10.2011 20:00:00.000

30.03.2012 21:00:00.000

28.10.2012 20:00:00.000

27.10.2013 20:00:00.000

I thought this problem was related to DST entering in action, but in 2014 the clocks
have been turned back on November 2, not October 26.

Has anybody an explanation for these Sunday candles?

I have found the problem. The timestamp in Dukascopy's OHLC data is not exactly UTC. It is a strange mixture of GMT and London-observed time. In the dates in which London moves to BST (late March and late October) there will be a 1 hour jump, there are apparent duplicated hours and skipped hours. In the later days the clock goes back to GMT. BUT, and this is the weird part, the time of the timestamp is not BST until the end of the summer time zone, but only in the switching days.

It seems that the tick data does not have this problem and contains proper UTC times.