Farrukh finds that 0.3% of the photon events failed
the v4.8.3 trigger
Beate's findings
I looked at the 0.7% of the photon events which did not fire
the photon25_iso trigger.
All of failures (22 events) don't have CES clusters, thus
are supposed to fail.
It turns out that the photon sample contains events
that failed the v4.3 PHOTON20_ISO trigger.
  CLOSE TO BE DONE!
d0 sample
(Trigger Path: L3_TEST_D0_471_DECOUPLED_v2)
Farrukh finds that 6.9% of the D0 events (that passed L3
v4.3) failed L3 4.8.3
Rolf's findings
In 4.8.3 there are more SVT tracks that do
not match a COT track, about 3%, which would
match with the 6% event loss observed by Farrukh
(3% comes in twice, for two tracks).
I did not find a particular part of the phase space
where this happens.
I do see some subtle differences between the 4.3.2
tracks and the 4.8.3 tracks:
The average chisq is a little worse (93.7 vs 88.3),
even though the number of axial hits is lower
(42.4 vs 43.9). I had expected the COT chisquare
to go down when >5 sigma hits get rejected.
A sine-cosine fit to the d0/phi distribution finds
the beamspot different by 1 mm (y=0.400 in 4.3.2;
y=0.495 in 4.8.3)
Chris and Avi confirmed there were some tracking changes
(for example taking into account the displaced beam center
for histogram tracking.) I looked at the distributions of
the 306 events that do not satisfy B_D0 in 4.8.3:
The L3 filter was 'right' to reject these events,
based on the input it had (SVT and L3 tracks).
So the differences are due to L3 tracking,
not the L3TwoTrackFilter.
The pi-K invariant mass distribution does show
a hint of a D0 peak but with a smaller yield than
the accepted events.
(nD0/nevt = 1/2 for accepted events, 1/4 for rejected)
Rejected events have tracks accumulating in the
phi=0.2 and pi+0.2 range, i.e. in the regions
where the beam-offset is maximal
So it appears as if at least half the loss (3%) is due
to improved fake rejection, and the other half might be
loss of signal. The present sample (raw B_D0 triggers)
is not ideal to quantify this, signal loss is better
measured on a sample of dead-clean D*->D0 pi, away
from trigger thresholds. Is anybody else seeing something
similar on the other samples?
  NEED TO BE DONE
Z -> ee sample
Trigger Paths:
L3_TEST_ELECTRON18_TIGHT_471_DECOUPLED_v1,
L3_TEST_PHOTON25_ISO_471_DECOUPLED_v1
Status: in the process of making an output file
with v4.8.3 L3.
W -> enu sample
Trigger Paths:
L3_TEST_ELECTRON18_TIGHT_471_DECOUPLED_v1,
L3_TEST_PHOTON25_ISO_471_DECOUPLED_v1
Status: in the process of making an output file
with v4.8.3 L3.
Muon sample
(Trigger Path: L3_TEST_MUON_CMUP18_471_DECOUPLED_v1)
Status: in the process of making an output file
with v4.8.3 L3.