I've finally tracked down enough of the remaining errors (there might still be one responsible for those peeks - I'll look into it as soon as I can) that I can calculate the luminosity by run:

This is calculated by taking the number of L2W accepts, subtracting the number expected to come from beam background (calculated as the sum of the average event rate in the two abort gaps, multiplied by an assumed 120 bunch crossings), then scaling by the prescale factor.  This yields the true number of BHT3 events that occurred in that run, which is then divided by the fraction of the detector that was live and then the cross section for BHT3.  I'll post more details after the meeting.