I have generated a new FIR filter bank for Hanford and tested it by running gstlal_compute_strain in partial calibration mode over 1024 seconds of data during a recent lock stretch, starting at 2016-09-27 at noon UTC (GPS second 1158926417). Kappas and strain spectra look reasonably sane compared to CAL-DELTAL_EXTERNAL_DQ. N.B., the EPICs values recorded in this new filters file are not correct, so kappas ought to be computed based on what is in the raw frames. There also appears to be an error with relative timing of the residual and control chains which is not yet resolved at either site; see the CALCS vs. GDS residual plot below and notice the large downward spike near 38 Hz. Aaron also noted this problem at LLO; see https://alog.ligo-la.caltech.edu/aLOG/index.php?callRep=28277.

The new filters file can be found in the calibration SVN:
aligocalibration/trunk/Runs/PreER10/GDSFilters/H1GDS_ 1158989379.npz

These filters were made using revision #3358 of the calibration SVN. 

Several plots are attached below:
(1) The h(t) spectrum as calibrated in partial mode using the new filter bank, and as read out of the front-end CALCS model and de-whitened;
(2) The ASD residual of CALCS vs. GDS (data calibrated using the new filter bank are of frame type TEST); and
(3) Plots of the new control and residual correction filters.

For plots (1) and (2), I had to highpass filter above a corner frequency of 8 Hz and use a 32-second FFT (with 16 second overlap) to get rid of some nasty spectral leakage. I filtered both channels identically (except for an extra de-whitening filter applied to CAL-DELTAL_EXTERNAL_DQ) and used a Kaiser window to overcome dynamic range issues. Plots of the new filters were made using the matlab code in the calibration SVN.