here is the feedback of the 1D pipeline analysis of the latest weekly run (dated December 14), which contains again the 3 arms. 

Among the 400 galaxies, 383 are classified as galaxies, and 17 as quasars. 

The final redshift (either from galaxy or QSO model) is correct (within 2e-3 in relative redshift error, which corresponds to 3 sigmas of the measured errors distribution) for 375 objects (93,75% success rate). Among the 25 failures 

10 come from objects with a featureless spectrum (which means that 97.5% is the top level of success we can expect)

7 come from PIPE1D errors (mainly on the z>5 objects with a H-alpha/Lyman-alpha mismatch). We will work on this

The remaining 8 show unflagged bad pixels. The list is in the attached file 

Among the 13 quasars, all are classified as quasars, and the redshifts are all OK (with a larger dispersion than for galaxies though, but definitely acceptable)

Among the 421 stars, 291 are classified as stars :

the measured recession velocity is exactly equal to zero, I suspect a sampling issue. We'll check that

130 stars are wrongly classified, 

93 as galaxies, but 76 with redshift measured < 0.05. Can therefore clearly be identified as stars. We'll work on the classifier though. 

We are left with 37 stars classified as QSO (all with redshifts > 3, hence wrong) and  17 as galaxies with high redshifts, which gives 8.7% of failures. The first inspection shows they all are K/M types with extremely deep and broad absorptions, I will check the templates.