The forms that survive transformation.
An eigenform is what a transformation cannot change. Pass the shape through and the shape returns. We take the name as a posture, not a promise. Under every learned representation, a few invariants do the real work. Most of what goes wrong in AI systems starts when someone mistakes the coordinates for the form.
An audit of the cosine gates that agent frameworks trust as quality checks. We built a corpus of paired texts that reverse meaning but keep surface form, then measured what the gates approve. The result is not subtle. Corpus, harness, and frozen results reproduce offline in one command.
Paper in submission. Code and corpora open with it.
A direction, not a claim. Past vector space, toward the symmetries that bind representations across scale. The question stays fixed while the systems grow: what survives when everything else transforms away.
Scott E. Frias
Creator and founder, Eigenforma. Co‑founder, Freemind Labs.