Per an email pointing this out, if test_deterministic.glm is run, the "unscheduled event" of the fuse blowing at 03:00:00 double counts the repair time of 7 minutes (it takes 14 minutes to replace). It appears that fuse.cpp is adding the repair time into the restoration time, then the eventgen.cpp event code is adding it in a second time. Not sure if anything beyond fuse has this issue, but it is something we'll want to fix.
Changes will revolve around code at 889:890 in eventgen.cpp (2.3) and how things are passed into 565 in fuse.cpp (2.3)
Diff:
Fixed and committed as part of r4820. Fuses were the only one expressing the problem, due to the "double timing" sets included in them.