Loop variable quantum gravity, causal sets or nets, causal triangulation theories and related ideas are themselves I think constraint systems. For instance, loop variable theory is a spinor, or spinor field, form of the 3 space plus 1 time form of relativity which has the Hamiltonian constraint NH = 0 and the momentum constraint N^iH_i = 0. The Wheeler DeWitt equation HΨ[g] = 0 is a canonical quantization form of the Hamiltonian constraint, and loop variable theory is a spinor form of this type of theory. The vanishing of the Hamiltonian NH = 0 or the canonical quantization HΨ[g] = 0 is due to the fact the manifold is global and there is no boundary from which one can compute with Gauss' law the mass-energy of the entire spacetime, or universe. It further have to be pointed out that loop quantum gravity (LQG) has yet to compute a one loop diagram properly. The reason is that if you have HΨ[g] = 0 it means you have ∂Ψ[g]/∂t = 0, and there is no dynamics! Computing a scattering amplitude, particles in --- > process --- > particles out, on a T-channel is not possible.
What do these constrain? Frankly I think they constrain string theory. Ed Witten has found that within string/M-theory there is a form of twistor theory. This is the so called twistor "mini-revolution" that started a few years ago. Twistors are in some ways related to loop variables, but they have more spacetime content. I think this segues into thinking about these non-string approaches to quantum gravity as constraint systems. String theory uses a "time" τ that is a string time parameter along the string world sheet with the parameter σ along spatial extent of the string. This permits one to construct Hamiltonians of the form
H = (T/2)[(∂X/∂τ)^2 (∂X/∂σ)^2], T = string tension,
for X the string variable. This contrasts of course with the LQG which has no explicit concept of time, because there is no way to define mass-energy in a global context. A Hamiltonian is the generator of time translations, which means the energy defined by the Hamiltonian is conserve (Noether's theorem). In string theory this corresponds to level matching of string modes, but in LQG E = 0 and there is no time translation. However, the spacetime is a target map from the string, which should correspond to HΨ[g] = 0 for a wave functional over the spacetime metric. LQG is then some type of constraint system.
There are also some interesting possibilities for duality principles. Barbour and Alves have proposed a form of shape dynamics, which is a symmetrical theory. The spatial relationships between elements that define a shape in space are symmetrical. Causal sets involve asymmetrical relationships between nodes that are connected by lines or into graphs. These represent temporal ordering. The two approaches seem to represent something similar to Penrose's tensor space of symmetric and anti-symmetric tensors in a type of duality. The duality in some work by Sparling and others is supersymmetry. I then conjecture that the correspondence between shape dynamics and causal nets (sets) is then a form of this duality or is a categorical equivalence to SUSY. This may then be another form of constraint, in particular the SUSY structure which exists in the AdS_n ~ CFT_{n-1} correspondence.
Cheers LC