Custom evaluation frames about winterfell

[4l0n50]

Just to complement, what you're saying is done here in the Cairo solidity verifier, which checks that OOD evaluations satisfy for example that (T_i(x) - T_i(g^33158*z))/(x - g^33158*z) is low degree. To what you said I would just add that you also need to modify the compose_registers method of DeepComposer struct.
However, I don't think it is a good idea to just define larger windows because not all values will be used. In the example above it will require a window of size at least 33158! I think one might instead just define which other (arbitrary?) cells might be needed when evaluating the constraints.

[maxgillett]

Yes, the frame will just provide a view of the trace, which shouldn't impact performance. Good point about the DeepComposer modifications. It would be wasteful to construct an OOD frame for all of the row offsets up to the window size (which could be quite large, as you demonstrate). One could either define in EvaluationFrame which row offsets are used in the constraints, or as in the StarkEx contract use OOD evaluations that are both column and row specific. The latter could be achieved by dynamically tracking relative trace indices during constraint evaluation, and storing these in a sparse matrix data structure. I think these are implementation details that can be abstracted away in the definition of EvaluationFrame.

[4l0n50]

However, I don't think it is a good idea to just define larger windows because not all values will be used.

Now that I think, it doesn't really matter even if EvaluationFrame depends on the whole trace. As long as your constraints doesn't depend on too many variables, everything should be ok