As I mentioned in the comments, I performed a similar analysis with variable transmittance schedule and noticed the same issue, so I posted this question. Within the question, I referred to 2 GitHub posts on the same issue but I am not sure if the issue is resolved!

One work around that I can think of is to avoid using EMS since, it seems, is not working properly with transmittance schedules.

• Instead, you can post-process your results by running the simulation once and reporting the corresponding output variables for the parameters used in the EMS control ruleset (i.e., solar radiation and operative temp in your case) and then come up with a transmittance schedule for every time-step of your simulation based on your control logic (assuming hourly analysis, you will have 8760 values for the transmittance schedule). Then you can use this schedule as an input for your model. Just in case you are running your simulation on hourly bases, you can use Schedule:File object to easily import your schedule from a .CSV file.
• A simpler approach: since you mentioned that you were able to see that EMS is modifying your transmittance schedule then you can use this modified schedule directly as an input.

So Basically, the idea is to use the transmittance schedule directly considering the control logic rather than relying on EMS to do the job!