Can all the TDT variants adopt a common strategy, but with different execution results, depending on source-code self-inspection and sim-inspection?

This sounds equivalent to asking "can a turing machine generate non-deterministically random numbers?" Unless you're thinking about coding TDT agents one at a time and setting some constant differently in each one.