Main idea

Once every card cited exists and no link points into the void any more, the set looks finished. It isn't: a set of cards that only cite each other closes its network long before it has covered the product. The network closed while only a minority of the modules had produced a card, and the coverage check run afterwards surfaced two central business objects that nothing in the network was calling for.

The closure check remains a good one, provided you know what it proves: that the set holds together. It says nothing about what never made it in, because it has no knowledge whatsoever of the thing being described — it knows only the cards.

Two checks, then, and never just one. That the set holds together is verified by looking at the cards against each other. That it is complete is verified by enumerating the product.

Why it matters

This is the most expensive confusion of its kind, because the check that fails to measure is also the one that fires by itself: unresolved links jump out at you, whereas an object never cited makes itself felt nowhere.

It transposes unchanged beyond cards. A test suite that passes proves the code under test behaves as expected, not that you tested the right things; a glossary whose cross-references all resolve says nothing about the words that aren't in it.

Nuances and limits

The distinction only holds if the thing described can be enumerated by some other route. On material with no possible inventory — a field of thought, an open corpus — coverage has no denominator, and only the cohesion check remains available.

And cohesion keeps a value of its own: a complete but incoherent set is no more usable than a coherent but incomplete one.

Open questions

• On material with no possible inventory, what replaces the coverage check without falling back on an impression of completeness?