The core

Everything up to here is about documents. This part is about the machine.

The decision is one sentence: one Rust core, one CLI, and thin R and Python packages that invoke it as a subprocess. No FFI. That is the architecture the sibling projects already use, so the largest cost of it — shipping Rust to CRAN — has been paid once in this household already.

The two chapters are the decision and the check on it. Architecture is why a core, why a subprocess rather than a binding, and the one place the arrangement strains. Parity is the part that makes any of it believable: the Python prototype is not replaced by the port, it becomes the oracle, and the port is finished when the difference between them is empty.

That distinction is the whole discipline of a rewrite. A port that improves what it is copying cannot be verified against it, because the improvement and the regression arrive as the same diff.