The five-question diagnostic is applied after the telic unit is chosen. Virus-plus-host-during-replication, thermostat-plus-room, and slow-timescale ecosystem are the units that pass. Membership is therefore a function of which loop is allowed, not of the questions.

Naming a unit before the test is required, not a loophole. The wall device fails detection and correction without HVAC and room. Convenient small boundaries are supposed to get a no; that is what the five questions are for.

Virus-plus-host-during-replication is chosen so hijacked host machinery counts as correction; thermostat-plus-room is chosen so HVAC counts. Every interesting no then looks like the wrong loop rather than a unit the test forbids.

Two readers run the same diagnostic only if they pick the same loop-plus-substrate pair before answering, and get the same yes or no. Different pairs with coherent five-answer sheets mean "telic" names the pair already chosen.

The language no-row wins by testing "the language itself," which was never the live candidate. Hypothetical: a speech community that shames, schools, and prestige-ranks, with no academy, has a norm, detects deviation, and corrects, and is closer to the academy row than to language-as-convention.

Civilizational telicity is withdrawn exactly when boundaries, specifications, or corrective machinery are contested — the stretches where capture, repair, and collapse tools would be used. The category idles when the later toolkit is needed.

Scoring parts (this agency telic, that crowd not) saves the toolkit and recodes a civilization-level no as "don't run the test on the composite." It does not save the opener, which lists civilizations in the shared class that the later section makes conditional.

A hurricane re-forms its vortex under a lasting gradient, as a thermostat returns to setpoint. Both return. The page's compare-versus-relax cut has no measurement in the storm that is not already the classification.

A hurricane does not compare itself to a genome. The specification has to be a template sitting in the system, not a restoring force. Pressure and temperature gradients are not that template.

The thermostat's setpoint is a voltage set from outside. If "carried" means the system generated the template, the thermostat fails the clean example. If it means present as a parameter, a river's bed geometry carries a channel.

Viability is not the same predicate across rows. Bacterium dies on shock; central bank loses legitimacy on a sustained miss; thermostat-plus-room goes to ambient while the wall box remains. Legitimacy-loss lets institutions pass on reputation; physical destruction would fail the central-bank row.

If the telos is the attractor revealed by correction, the specification is inferred from what the system protects, not read off in advance. That inference is also available for a hurricane's vortex.

Membership and slogan-versus-revealed are different jobs. Membership uses whatever spec the machinery is carrying. Revealed-versus-stated is a later check for systems already in the class. A university can be telic under its machinery and still lie on the website.

Membership still needs a specification written down before perturbation. For the university, truth, prestige, or "whatever correction will reveal" are different slot-fills. The third makes membership use the later check.

Cancer, capture, and mesa-optimization are the same failure only if the real telos is the revealed one. If membership used the stated spec, capture would be missing the spec, not installing a new one. Same-architecture needs one of those diagnoses.

Literal identity of telicity across a bacterium and an institution is a match of five abstract headings. A cell's correction is chemistry; a court's is procedure plus people who can ignore the procedure. Matching headings is a shared outline, not the same formal sense.

The claim is identity of jobs — detect, correct, stay viable — not identity of physics. Different carriers, same roles. That is what "not by analogy" is supposed to buy.

Cancer-failure is correction machinery eating the capacity to keep correcting. In a cell that is metabolism. In a ministry, reporting can keep "correcting" on paper while caseload worsens. If extra facts (whether the report binds, whether anyone can stop it) are still required, the five properties did not carry the failure.

Wiener already treated feedback across animal and machine. The named extra is a membership test that lets the later tools travel. If a yes still needs extra institutional facts, the extra shrinks to a vocabulary for systems already wanted as comparable.

The page splits training loop from deployed policy and says a bare next-token model is not telic. A frozen deployed policy has no reward signal and no comparison to the training spec, so detection is gone unless the scaffold is put back. "The trained policy in a deployment environment is one telic system" may fail the detection question on that split.