shows

Reads subject off the binding's public value surface — "votes" in it.keys, it.contains("x"), it.store.values.isNotEmpty(). Must be a function of the value, not of object identity; the suite checks that by evaluating it on a freshly joined equal state.

It must also be a function of the SUBJECT, and that half the suite cannot check. A predicate written in terms of the sibling samples — { it != retired }, { it == asserted || it == reAsserted } — reads true → false → true, is a function of the value, and satisfies the freshly-joined check too, since retired.piece(asserted) == retired by the ascent the guard has already established. It passes every arm and carries no information at all. So: name a subject and read that, and never write this predicate in terms of asserted, retired or reAsserted.

This is #2157's own lesson one level up. Naming the three states is not evidence, which is why shows exists; naming a predicate is not evidence either unless the predicate is about a subject rather than about the states. Closing it structurally would mean inspecting a lambda, which the suite cannot do — SiblingReferencingShowsRigTest in this module's commonTest pins the limit instead, by being a green binding whose predicate is exactly the one above, on a lattice with no retirement in it at all (#2184).

A one-dimensional subject space is not the same defect. Causal<DotSet> and Causal<DotFun<V>> have exactly one thing to show — membership, or "the register has a value" — so it.store.dots.isNotEmpty() and it.store.values.isNotEmpty() are correct there. Those are about a subject; it simply has no key to name. The rule is what the predicate is written in terms of, not how many keys it mentions.