We have had a persistent feature in the track-by-track QA plots.  The global track efficiencies for STV tracks exhibit a significant DIP between (roughly) -120 < phi < -60, as illustrated by this efficiency plot from 9/16.

Figure 1 -- Drop in STV efficiency as a function of phi 9/16.

For the class of tracks where STI and STV have significant overlap in hits (aka "matched" tracks", with more than 10 common hits) we can plot the difference in Nfit points vs phi.

Figure 2 -- Difference in Nfit points between matched STV and STI tracks.

There is a clear "feature" sitting at about Δ Nfit = -8.   This looks suspicously like an RDO, as we know there are bad ones in this region.  Friday, Victor and I talked about possible sources.  He identified a "STOP TRACKING" condition which exists in STV but doesn't exist in STI.  It is a condition triggered by the size of the uncertainty in the propagated position of the track.  If it grows too large, no new hits will be added to the track.  This is in place to stop the tracker from adding hits which are clearly unassociated.  But the possible side effect is to reduce the number of fit points, which would then reduce efficiency relative to STI.

We increased the threshold at which the STOP TRACKING condition is triggered, and see improved results.  The following plots compare the difference in fit points vs phi, and log10(P) for the 9/16 version of STV and the 9/23 version with the increased cut.  Note that production of the 9/23 data sample is ongoing, and this analysis is based on a smaller sample of data.

Figure 3 -- Δ Nfit vs phi for 9/16 on the left, and 9/23 (larger threshold) on the right.

Restricting ourselves to sector 8 and looking at the plot vs log(P) ...

Figure 4 -- Δ Nfit vs log(P) in sector 8 for 9/16 on the left, and 9/23 (larger threshold) on the right.


Increasing the threshold where the stop tracking condition is applied has clearly improved the agreement between STV and STI in the number of fit points on matched tracks.  Once we have more statistics available, I'll run the full track-by-track QA suite.  Based on these results I expect that to show improved tracking efficiency in STV.