Yes. There is a bunch of people who started with ST instead of WW/CE, where it was much more obvious. The CE and HF AI sort of teaches it: it is merciless at it in many situations.
I’m doing a little frame data investigation in Hyper Fighting at the moment (putting a little work into the wiki while I’m at it), and noticed a couple of things which agree with/add to this data.
I’m using mame-rr with the lua scripting as described at http://code.google.com/p/macrolua/wiki/FrameDataExamples , with the MAME cheat code for SF2HF “frame skipping turned off” (this is crucial, but also not necessarily bulletproof). My results are consistent, so I don’t believe there’s any randomness going on.
For all the following I’m using Vega (Claw) mirror match.
Theory:
Frame data suggests +1 extra advantage on block
Script evidence:
claw s.lp (frame advantage test) = +5
#W10 1.W25,U.W4,-U.! #neither jump
#W10 1.W26,U.W4,-U.! #both jump
claw s.lp on block (frame advantage test) = +6
#W10 -_R,+1.W25,U.W4,-U.! #neither jump
#W10 -_R,+1.W26,^R,U.W4,-U.! #p1 jumps
#W10 -_R,+1.W26,^R,U.W5,-U.! #both jump (P2 requires an extra frame to successfully jump)
This agrees with the first post findings that a blocked s.MP causes 1 more recovery frame than a s.MP on hit.
Theory:
Hitstun deteriorates by 1 frame after the initial attack
Script evidence:
#claw combo: cr.lp cr.mp (cr.lp is +6, cr.mp has 5f startup, making this a 1 frame link)
#W10 _D-_D,+1.W6,-_R,W19,+2.W29.! #works
#claw test combo: cr.lp cr.lp cr.mp (+6 into 4f, +6 into 5f, i.e. 2f link followed by 1f link [reduced to 0f link after the second hit])
#W10 _D-_D,+1.W6,-_R,W19,+1.W25,+2.W29.! #cr.mp is blocked
#claw test combo: cr.lp cr.lp cr.lp (2f link, 2f link [now 1f after the second hit])
#W10 _D-_D,+1.W6,-_R,W19,+1.W25,+1.W29.! #works!
#W10 _D-_D,+1.W6,-_R,W19,+1.W26,+1.W29.! #3rd cr.lp is blocked
This also agrees with the first post - recovery for the hittee is one frame earlier after the second attack in a combo.
Theory:
Later hits in a combo may deteriorate hitstun further
Script evidence:
#claw test combo: cr.lp cr.lp cr.lp cr.lp cr.lp (2f, 1f, 1f, 1f)
#W10 _D-_D,+1.W6,-_R,W19,+1.W25,+1.W25,+1.W25,+1,W15! #5th cr.lp is blocked
This test is a little controversial, as this could potentially be a hitbox issue - although I’ve run the macro up till the last hit then switched to the hitbox script (can only have one script running at a time!) and it looks like the boxes still overlap. Further suggestions for experiments welcome.