I looked at the rates & data sizes by time in the current production running (run  26168017).    We see bifurcation of the rates in the high level rate plots in my plots:

I was worried that this was an effect either of background or of some sort of issue within the DAQ.   I looked at the BX for each event taken and resampled both rates and event sizes for various time bins:

And the zoomed versions

The conclusion here is that there are no backgrounds leaking to this bifurcation.    We do see a lot of gaps on the order of 1-2 seconds, and these lead to smaller rates in the less detailed plots.     The 1-2 second deadtimes are TPX/TPC deadtimes.    This is coming from the RDOs, but I don't know if this is fast autocorrections, or whether they are due to blocking when front end buffers are filled.