I used Tony and I's statecounter.py to take a look at the past year of data using minute trends. Minute and second trends do some rounding that I had to consider in my search, ETMX goes from -8.9 to +6.05425 during a lock aquisition, these average to ~ -1.42. I ended up searching the chan H1:SUS-ETMX_L3_LOCK_BIAS_OFFSET for above and below -2.0 doing the following calls:
This gave me an outfile.txt file full of results in the format "idx (data_idx_start, data_idx_stop) gpsstart gpsstop duration" which I did some brief analysis on yielding:
Percentage of the time the Bias offset was [+]: 43.70 % *Past locked LOWNOISE_ESD_ETMY
Percentage of the time the Bias offset was [-]: 56.25 % *Between PREP_FOR_LOCKING and LOWNOISE_ESD_ETMY
Total duration in [+]: 205.3 [Days], [-]: 159.5 [Days] over 365.0 [Days]
Data timespan missing due to minute rounding and FW restarts 0.1965 [Days] (0.054 % of total time).
Where the BIAS_OFFSET is changed from negative to positive has changed a little over the past year but it's always during one of the final states of ISC_LOCK and it's always reset to negative PREP_FOR_LOCKING.
I haven't done the same search for ETMY but looking through where it's set in ISC_LOCK and ndscope I can say that ETMY spends the large majority (>90%) of time at -4.9. Specifically it stays at this value for 4min 23sec per acquisition, it gets changed to +9.3 during LOWNOISE_ESD_ETMY but it's brought back to -4.9 in the next state LOWNOISE_ESD_ETMX as we switch back.