Where do we get the info we need for TMS?

[moradology]

Interpreting TMS seems to require that we make some guarantees about coextension of computed sources. How, then, is such information provided? Some options:

1. make the AST bigger and more complicated so that it can inform the interpreter of such things (this seems bad)
2. assume the provider of sources is only providing tiles for apples to apples comparisons (dangerous, though currently the strategy of RFML)
3. Create new instances of the interpreter with expectations for output and make sure that tile sources provide enough information for the interpreter to fulfill said expectations (this is likely the path)