Did you get this to work? I would suggest changing the time range of your calculation in line with the dates your archives stop and start, in case of any overlap.
eg. If the calc is from Jan - Feb, but there was an archive shift in the middle of jan, try changing it so it only calculates against one archive file.
If it works in one file and not another then you can probably narrow it down to a problem with a single archive file.
It may be worth reprocessing all archives over the time period of this calculation, as "record header" errors are usually a sign of archive corruption. Has anything changed with the tags recently which may not be reflected in older archives? (range, point type etc?)
Who ate all the PIs?