These were exactly my thoughts on this matter. So, let’s summarize the problem and the possible solutions.
The problem is that when an observation is scheduled early it uses the current TLE which could be changed until the time that observation will run.
The easiest solution would be to keep the same window and just change the TLE in order to get what we will manage to get but with the right frequency. I think that this should be implemented in the first stage.
The hard solution is to re-calculate the pass and set new start and end time for the observation. This is hard because in the same station there could be other scheduled observation and that could create overlapped observations.
In order to avoid this we should extend the scheduled time by some minutes/seconds? before/after?. This way we will be able to change easier the observation without any overlap.
I think that for this we will need some data from an analysis on how much start and end time is affected by changed TLE. As @cgbsat noted for satellites without correction of their orbit we shouldn’t expect any big change, but for the other ones like ISS and NOAA maybe is big enough.
I’m very interested in such an analysis, I’ll try to perform one the following week if I spare some time.
Another aspect is that while scheduling is done manually we want to be able to schedule for a couple of days. However when auto-scheduling is ready, the scheduling would be done a couple of hours before the observation, increasing the chances that the station has the right TLE.
Anyone else? Any thoughts?