Thanks for the clarification. What about deactivating secondary layer for digital twin? In some situations it might automatically switch back on though, if otherwise your training plan request can’t be accommodated by your model alone. For example if the volume you’re requesting is much larger than what you have trained in the past.
There’s also the new/improved bulk fit file upload: Bulk .FIT file upload: import your full training history which I think you should give all the data you have. Maybe your own data data without secondary layer would then be enough.
Maybe what chat means is you could optimize a different metric than strictly marathon time. For example Endurance Power + Critical Power template if you’re using running power. Or Endurance Pace. That could also be an option.