InteractionKit
Typed experiment infrastructure for testing whether corrective interfaces must match how AI communication fails.
Interactive evidence with bounded claims. Each artifact can be operated in the browser, and each carries its own claim boundary: what it demonstrates, what it does not, and where the executable evidence lives.
These are not product demos and not validated results. They are the inspectable layer of one research program: the same systems described under Research, opened up for direct inspection.
Typed experiment infrastructure for testing whether corrective interfaces must match how AI communication fails.
A provenance-preserving prototype for comparing candidate-revision strategies on a fixed constructed corpus.
A small single-process evaluation runtime that makes execution attempts, ambiguous failures, and report denominators separately inspectable, under a machine-readable reliability contract.
Portfolio rule
An interactive artifact is not a validated scientific result. Software conformance, constructed-corpus outcomes, simulation behavior, and human evidence answer different questions; none is used as a substitute for another. Every claim on these pages points to the evidence that bounds it.