Unexpected behavior generating toys from model with 2 POI

[jennetd]

Reporting an issue seen with CMSSW_10_2_13, combine v8.2.0

I am running an analysis with 2 POI (rggF and rVBF). I submitted 100 parallel jobs of 10 toys each, expecting 1000 toys as a result. Instead, I see exactly 4x the expected number of toys in the output tree. The output tree contains sets of 4 toys with identical POI fitted values and errors, but with different values in the "limit" branch. In one of the four toys in the set, the limit branch matches the fitted value of the first listed POI. I attach a few plots here for demonstration, and in case it is useful I have uploaded an example output file on lxplus here:
/afs/cern.ch/user/j/jdickins/public/combine-issue/bias20VBF.root

limit_rggF.pdf
limit_rVBF.pdf
limitErr_rggFErr.pdf
limitErr_rVBFErr.pdf

This is the command I used for toy submission:
combineTool.py -M FitDiagnostics --setParameters rVBF=$bias,rggF=1 --trackParameters rggF,rVBF --trackErrors rggF,rVBF -n bias${bias}VBF -d ${modelfile} --cminDefaultMinimizerStrategy 0 --robustFit=1 -t 10 -s 1:100:1 --job-mode condor --task-name VBF$bias

The issue is two-fold:

• The output should not contain 4x the number of toys requested
• The limit branch should contain some sensible value for models where 2 POI are fitted simultaneously. It appears somewhat random at the moment.

Thanks, and let me know if any additional information from me can be useful.