Looked at the MinuitVF performance in 2014 AuAu events using the available MuDsts on NFS.  I think this was a test production?

The specifics are the production is AuAu_200_production_low_2014 with P15ic.  No trigger selections, this can influence vertex finding efficiency but for the moment is a 2nd order effect.  The last time the ranking was tuned was for 2007 AuAu 200.  I had spent some time with it for the BES data, where some of the same issues were there.

Namely in the efforts to not penalize vertices with a small number of tracks, they are given a higher rank than they ought to.  This is difficult, when there are only a few tracks assigned a rank will always suffer from the statistics.  In the BES data, we divised a series of cuts based on the TOF multiplicity, to require that the vertex multiplicity was correlated with the vertex multiplicity.  This is because the vpd is very inefficient there, and can't be used as a criteria.  At that point, looking at the vpd Vz vs index 0 vertex, it was decided that |vpd Vz - Vz0| < 3 cm could do the same thing in AuAu 200 GeV.  (6 cm works as well, the population only differs by ~1%).  The intention was to add the TOF to the vertex ranking.....

The TOF ranking was added to PPV, used for pp, but somehow not in MinuitVF.  This should be done as well.

Looking at 42,553 events, the % with both a reconstructed vertex and a vpd Vz was 95%.
The percentage that were "bad" is about 11% (i.e. 95%x11% of events).
Only 10% of these events had the next highest ranked vertex within 6 cm of the vpd vz.
Spinning through the rest of the vertices to find the closet vertex to the vpd, I found that the index could be nearly anything.  But, the multiplicity of this vertex is generally correlated with the tof multiplicity.  I call these "best" vertices in the presentation.  The rank of these vertices is poor....  I think if the tof is added, the ranking will be higher and we will see a better correlation between the vpd vz and index 0 vertex.  Then perhaps the other parameters can be improved.