Also keep in mind that...

• you need to synchronize this "internal" posture state with the animations that are happening; edges between states will have corresponding animation sequences to initiate.
• you need edges for when the user *fails* to press certain buttons within the valid window; what state and animation is next if they don't press the next Jab button in your planned combo?
• you need reaction edges; what happens if you are hit by various classes of attack from the opponent at certain times?

The final form of your graph needs to be easy for the engine to consume. To those ends, you may want to make your graph edges simpler; each state has an animation which it starts, and has edges leading out; one of those edges will be "timed out", which leads to another state which implements the next time window or some kind of follow-through. Some of those edges will be interrupts, such as being struck by various attacks by the opponent, and they direct you into guard-window states (a split second to counter) or stun states (an I've-been-hit animation, with its own edges to leave it).

And all that on top of directional input... stepping back, holding down+forward while pressing punch, jumping... the more complex you want it, the more complex it needs to be.