The rule

Not assessable. The trade has no occupation code of its own, or its code was folded into a broader one. Any figure that appears against it belongs to the parent occupation and is not a count of this trade. No category is assigned, because none can honestly be computed.

Critically endangered, on either of two tests. A national base under 500 — below that point a training route on paper cannot replace the people leaving, whatever the paperwork says. Or a base under 2,000 with at most one registered sponsor nationally, which is no functioning route at all.

Endangered, if any of the following fire: a base under 10,000 with 30 or fewer registered sponsors nationally; a projected decline of 15 percent or worse over ten years; a base under 25,000 with 100 or fewer sponsors where postsecondary credentials have at least halved since 2000; or no institution awarding a credential in the trade at all.

Currently viable. Base and training route both holding. This is a statement about the next decade of supply, not about the health, pay or dignity of the work.

A registered sponsor is an organization permitted to train, not a person training. The rule treats sponsor counts as meaningful above a few hundred practitioners and meaningless below, on the reasoning that one retirement wave ends a very small trade whatever its paperwork says. A reader who disagrees should re-run the rule without the sponsor tests and see what moves.

The credential test is deliberately asymmetric. It is allowed to move a small trade and not a large one, because 820 sponsors is a working route however the college numbers went. That asymmetry is a judgment, and it is stated rather than hidden in a weighting.

Suppression is recorded, not corrected. Where a state count is withheld for confidentiality the trade is marked suppressed rather than imputed. Suppression is itself a finding: a trade too small to report is a trade too small to see.

The thresholds are round numbers. 500, 2,000, 10,000, 25,000, fifteen percent, half. They were chosen before the data was run and have not been moved since. Round numbers chosen in advance are more honest than precise ones chosen afterward.