I can understand some skepticism, but I am pretty certain that if you position your arm correctly from the start of the stab, to the end of the twist, that is plenty of digits to work with to collect that the player rotated the wrist, imo.
I’ll preface below with that, redesigning the hit-detection system (aside to include wrist turning, etc) is probably not necessary to implement a lot of new features. It would just be an improvement.
That said, part of this change suggests designing a more intelligent warrior system capable of “guessing” (within a narrow window) what the player is doing, rather than locking them out of combos for a bad-swing register. An ideal system won’t pull the trigger immediately, it will wait an unnoticeable amount of time (to human perception) until it’s determined the valid/intended swing. So as to not confuse for example the thrust/bleed combo, a large stabbing motion will be looked for and then it will start looking for large wrist turning, triggering the bleed attack. Whereas if the player thrusts/turns the wrist, and then hesitates(but the wrist does not rotate inward) or begins swinging again, then it will know the move was a thrust and a bleed is not going to follow up. As just one example. Another option is to look-back at the previous swing for possible results, in this instance. If a thrust is recorded, then your next swing-state should look back at it to see how much wrist-motion was present during that thrust. If the number is low, then it probably was not a thrust+turn. Thus, this current swing-state would be expecting a higher amount of wrist-motion to invoke bleed. Otherwise, if the last swing only registered as a black dot/thrust, but had 50% of the wrist motion required, and then this swing-state also contained ~50% of the required wrist motion, which began near the end of the last state, then we are most definitely performing a blackdot/wrist rotate and this swing-state is actually no different from the last, it was just a mis-fire. Something along the lines of that… This isn’t really my 100% right here lol, if I sat down and fleshed it out I could but they can figure out those details themselves.
Similar-ish to the runemage system, where a calculation is done at the end of a drawing, except this would be on the fly within small time-windows, triggered by swings “starting”(as they are done now). Once the system is like 90% sure a swing is an upswing, and not a right swing for example, it will exit the window and begin looking for the next swing. Otherwise, if the window closes, it will take the highest valid swing. Again, these time windows are so tiny that they would not affect your swings negatively.
They could even have an invisible “ghost blade” follow your strikes at some delayed elliptical speed(so that it never strays too far from your actual sword), and take the difference of motion(quaternions, for example) between your “two swords” to try and guess what you were really doing. Or store your motion information at some critical point during a swing. I have no idea what their system actually does.
Wikipedia is a garbage fire of overly complicated information, but quaternions could be useful here if they aren’t already using them. https://en.wikipedia.org/wiki/Quaternion