YES(?, b + 4)

Initial complexity problem:
1:	T:
		(?, 1)    eval(a, b) -> eval(a, a) [ 0 >= a /\ b = 1 ]
		(?, 1)    eval(a, b) -> eval(a, a) [ b >= 1 /\ b + 1 >= 0 /\ b >= a + 1 ]
		(?, 1)    eval(a, b) -> eval(a, 0) [ a >= 1 /\ b = 1 ]
		(?, 1)    eval(a, b) -> eval(a, b - 1) [ b >= 1 /\ b + 1 >= 0 /\ a >= b ]
		(1, 1)    start(a, b) -> eval(a, b)
	start location:	start
	leaf cost:	0

Repeatedly removing leaves of the complexity graph in problem 1 produces the following problem:
2:	T:
		(?, 1)    eval(a, b) -> eval(a, a) [ b >= 1 /\ b + 1 >= 0 /\ b >= a + 1 ]
		(?, 1)    eval(a, b) -> eval(a, b - 1) [ b >= 1 /\ b + 1 >= 0 /\ a >= b ]
		(1, 1)    start(a, b) -> eval(a, b)
	start location:	start
	leaf cost:	2

Repeatedly propagating knowledge in problem 2 produces the following problem:
3:	T:
		(1, 1)    eval(a, b) -> eval(a, a) [ b >= 1 /\ b + 1 >= 0 /\ b >= a + 1 ]
		(?, 1)    eval(a, b) -> eval(a, b - 1) [ b >= 1 /\ b + 1 >= 0 /\ a >= b ]
		(1, 1)    start(a, b) -> eval(a, b)
	start location:	start
	leaf cost:	2

A polynomial rank function with
	Pol(eval) = V_2
	Pol(start) = V_2
orients all transitions weakly and the transition
	eval(a, b) -> eval(a, b - 1) [ b >= 1 /\ b + 1 >= 0 /\ a >= b ]
strictly and produces the following problem:
4:	T:
		(1, 1)    eval(a, b) -> eval(a, a) [ b >= 1 /\ b + 1 >= 0 /\ b >= a + 1 ]
		(b, 1)    eval(a, b) -> eval(a, b - 1) [ b >= 1 /\ b + 1 >= 0 /\ a >= b ]
		(1, 1)    start(a, b) -> eval(a, b)
	start location:	start
	leaf cost:	2

Complexity upper bound b + 4

Time: 0.135 sec (SMT: 0.129 sec)