Duals should pair up with the opposite "axis" dual. I have understood this after observing many dual couples:
ILE-Ne - SEI-Fe
ILE-Ti - SEI-Si
LII-Ti - ESE-Si
LII-Ne- ESE-Fe
SLE-Se - IEI-Fe
SLE-Ti - IEI-Ni
LSI-Ti - EIE-Ni
LSI-Se - EIE-Fe
SEE-Se - ILI-Te
SEE-Fi - ILI-Ni
ESi-Fi - LIE-Ni
ESi-Se - LIE-Te
EII-Fi - LSE-Si
EII-Ne - LSE-Te
IEE-Ne - SLI-Te
IEE-Fi - SLI-Si
You're welcome to observe these pairings IRL to see that they are optimal. With this pairing, the compatibility is maximal, but they're also trickier to stablish as per the duality dynamic.
With the same axis dual, the ITR is a mixture of activity and semi-dual, which in itself is not "bad", but not the real deal duality. You can also fight with you same axis dual quite a bit.
Next part of the compatibility schematic; DCNH:
Dominant (D): N - Optimal
D- Bad
H- Good
C- Tricky
Creative (C): H - Optimal
D - Tricky
N - Bad
C - Very good
Harmonizing (H): C - Optimal
H - Good
N - Tricky
D - Good
Normalizing (N): D - Optimal
H - Tricky
C - Bad
N - Good
I wanted to add that DCNH compatibility is very, very important when pair-bonding, even to the point of making less-than-ideal ITR's socionic wise, pretty good (like for example, Contrary can seem much better if you match on the DCNH).