The Direct (and Recursion) method is very stable, but a bit slow, whereas Krylov can give instabilities, but it's very fast. If no numerical issues occur, they should all give the same numbers in the end (within some precision), but as you noted if the Krylov method gives negative transmission, one should switch.
It should be said that we have improved the Krylov method in 12.2 (and some additional fixes in 12.8.x) such that if negative results are discovered, the engine switches internally to Direct by itself, in order not to return unphysical values, and also the Krylov method itself has been made more stable. Thus we hope the problems with negative transmission should be more or less eliminated in the newer versions of ATK.
To return to the question - it's a bit unchartered territory. If you want to be on the absolutely safe side, you should redo the scf with Recursion. Pragmatically, however, as long as you get reasonable transmission with the mix of methods, I think you are quite ok. The negative transmission only occur for specific energies, after all, and you are somewhat unlikely to hit those on the contour. But don't hold me to that answer - it's your own responsibility to check
