Hi Santiago, OK, I have run it ~15 times without problems last time. I however just disabled the wrong of the two errors (one causes the other), so I switch this now.
However, after digging into the code, the failure looks mystic, since the number there is just created from a substring "0.9x11" by taking everything left of the x, and then divided by 3600. The string is fixed in the test. It looks like that there sometimes an additional digit slips in, but in one case (20161207T200220Z) the extraction results in 25000 instead of 0.00025. Really strange. But I don't feel that I can do much here (it is already forwarded). Upstream is informed, however. So, I disabled the (now hopefully right) test, and the package could be built now 40 times in a row without failure. Thumbs crossed :-) Best regards Ole