The date conversion assumes the current year & timezone, which leads to inappropriate conversions when the log parser and the log file are on different sides of year or DST boundaries. The most benign result of this will be a one-hour gap in the graphs as the time jumps. The least benign would be a whole year of unprocessed log information because the new timestamps are "older" than the measurement that came in at the end of December from the previous year.
Workaround: regenerate the scoreboard and RRD files on the first of the year.
Suggested fix: during the date conversion, check to see if the resulting date is in the future. If it is, then try to figure out if it should be last year or an hour earlier.
I have implemented the suggested fix and checked it it, but haven't tested it. Along with testing the makefile, getting this tested will be a big step toward releasing version 0.13.