I have tried to check what could be the physical origin of the shape of the glitches with ~25min occurence rate. In h(t) the glitch can be affected by the different filterings and combinations of signals used to build h(t), so it is hard to intepret.

Figure 1 shows compares the B1 Audio channel with and without the glitch, with several models, and cyan is the quadratic difference between the data with and without the glitch. The models involved the DARM loop (L in the legend). When a glitch occurs in a loop just before the sensor then the shaping of the glitch by the loop is 1/(1-L), when the glitch occurs just after the sensor it is L/(1-L), when a glitch happens somewhere else in the loop the one needs to take into account the response of the parts of the loop between the glitch and the sensor.

The best fit model looks to me to be the purple curve which is 1/(1-L)/F^3. This could be for example a glitch that is a step function (1/F in the frequency domain), that is on the current applied to the coils so before arriving to the photodiode it is filtered by the 1/F^2 of the suspension, to make in total a 1/F^3 shape.