I summarized yesterday's SDF issues.
Basically, the issue is the incomplete description of SDF-related actions in the document.
Sorry for the inconvenience.

The proper procedure for the worker creating the SDF diffs is as follows:
1. Accept SDF diffs at the observation state of the LSC_LOCK guardian, taking snapshots of the SDF screens just before acceptance.
2. Re-load the observation.snap file for the corresponding models to confirm the presence of numerical rounding errors. If SDF diffs appear during the reload cycle of the SDFs, there are likely numerical rounding errors, so revert the SDF in this step.
3. Relock the IFO to ensure that no SDF diffs appear during the relock cycle. If SDF differences appear during the relock cycle, they should be changed by the guardian. In this case, the SDF differences should be accepted.
4. If SDF differences appear in step 3, step 2 should be repeated to confirm that the numerical rounding errors do not exist. If exist, modify the guardian codes to avoid numerical rounding errors.

Initially, I thought steps 3 and 4 should only be done when the guardian codes are modified, but that is not true.
The GOOD value recode for OMMT1 uses full-scale values, but reverting to GOOD values at the RESET_ASC_FEEDBACK state rounds the GOOD values when inputting OFFSET values at the OPTICALIGN filter banks.
Therefore, even without changing the guardian codes, steps 3 and 4 should be performed if the OMMT1 alignment changes.