globe gets read/written backwards in its bandwidth graphs
https://atlas.torproject.org/#details/7552CA84FB125059DC2959A6BE01A6A8107B3523 shows a big jump in bandwidth, especially read bytes.
whereas https://globe.torproject.org/#/relay/7552CA84FB125059DC2959A6BE01A6A8107B3523 shows a big jump in bandwidth, but especially written bytes.
Examination of the extrainfo descriptor for the relay in question shows that it's read-history, not write-history, that has the big numbers.
So I think globe has its labels mixed up?