retirementIsMeaningful

Whether this type can retire — stop showing something it once showed, while still only moving up the lattice.

What counts as retiring is defined in exactly one place: us.tractat.kuilt.conformance.lattice.OpKind, as an op retires when it takes an observation back without putting another in its place. Read it there rather than re-deriving it here; the definition has drifted between surfaces twice already, which is why it now has a single home (#2146, #2159).

This surface reads it more generously than the lattice bindings do, deliberately. A register's set supersedes rather than withdraws, so MVRegisterConvergenceTest declares no retiring op at all — yet MVRegisterConformanceTest sets this true and names a set that stops the register showing an earlier value. OpKind carries the rule that makes both right: classify strictly where the answer is averaged, generously where it is checked. There the classification feeds a rate — the retirement vacuity floor — which a label every op carries would clear by construction. Here it gates a single constructed triple whose every step is asserted (samplesReAssertAfterRetirement), so a generous reading buys one more checked shape and can inflate nothing; and a supersession that did not really stop the old value being shown reds the guard rather than passing it.

A binding that sets this to true must also name the shape in retirementReAssertion. The false default is an opt-out, not evidence: a type that genuinely cannot retire reads false, and so does a retiring binding that has not named its triple yet (#2167). A binding that leaves it false on a type its lattice binding declares retiring should say why.