We do a series of short runs where one trigger is allowed to make events. This is done by making prescale factor low for that trigger and high for all others. Check if deadtime makes sense.
Run ... Trigger .... Prescale ... Raw Rate ... DAQ rate ... Deadtime
1850 ... 1 ......... 80 ......... 80 kHz ..... 100 Hz ...... 0 ... (ok)
1851 ... 2 ......... 1 .......... 120 Hz ..... 120 Hz ...... 0
1852 ... 3 ......... 14 ......... 1400 Hz .... 100 Hz ...... 0
1853 ... 4 ......... 6 .......... 6 kHz ...... 100 Hz ...... -75% (!?)
1854 ... 5 ......... 1 .......... 5 Hz ....... 5 Hz ........ 0
(Note, T5 rate is low, so need to measure over long time to get accuracy. datamon measures in 4 sec, I did it with >>4 sec here.)
1855 ... 3 ......... 14 ......... 1400 Hz .... 100 Hz ...... 0
Repeated T3 since beam was unstable in 1852 (I took my eye off beam).
Next, do a run with normal prescale factors for production except T4 is prescaled away. It was run 1856. Result: deadtime makes sense now. No longer negative. BTW, I also checked that the correction for T1,T3 correlation is correct, and analyzed a file to verify that the TS-accept scaler agrees with number of events in CODA file.
Therefore T4 is the sole culprit for "negative" deadtime. What is crucial is that the SAME signal that is plugged into trigger supervisor is also plugged into scalers. This seems to be the origin of the problem, we can check during an access. However, it would be correctable since THaNormAna class computes deadtime for each trigger separately. It should be verified that the deadtime is approx the same for each trigger.
A copy of this log entry has been emailed to: reitz@jlab.org,gayou@jlab.org,saw@jlab.org