The Scaffold Isn't the Building
A species page can have a taxobox, a binomial name, an authority citation, a country of origin, and a publication year, and still contain exactly one sentence about the organism itself. The page is complete. The moth is not.
Take a Costa Rican species in the family Blastobasidae, described by David Adamski in 2002 as part of a 147-page monograph on a related genus. The Wikipedia entry gives you genus, species, the describer's name, the year, the country. No wing pattern, no host plant, no mention of how many specimens were collected or where. Every field a taxobox asks for is filled in correctly. The actual biology, the stuff the taxonomy is supposed to be a handle on, is entirely absent. It's not even summarized badly. It's just not there.
This isn't sloppiness. It's revealing something structural: a record like this is being checked on two different axes, and they don't track each other. One axis is the scaffold, whether the name follows binomial convention, whether the citation resolves to a real monograph, whether the required fields are present and formatted right. The other axis is the substance, whether the thing the fields point at actually got described. A record can max out the first while scoring zero on the second, and nothing in the format stops that from happening, because nothing in the format is checking for it.
The reason this keeps happening at scale, rather than being a rare failure, comes down to cost. Scaffold-completeness is cheap to verify. A template or a bot can confirm a taxobox has every required field, that a DOI resolves, that a name matches the genus it claims to belong to. None of that requires knowing anything about the organism. Substance-completeness is expensive. Checking it means reading the monograph the citation points to, knowing enough about the genus to notice what's missing, and judging whether the one sentence given is adequate or just occupying the space where a description should be. That's slow, specialized labor, and there's no automated stand-in for it.
So most systems check the axis that's cheap, not because it's the one that matters but because it's the only one that scales. A citation gets validated by confirming the link works, not by confirming it supports the claim it's attached to. A compliance checklist gets signed off by confirming every box has an entry, not by confirming the practice behind the entry exists. A résumé gets screened by confirming every section is populated with plausible nouns, not by confirming any of them happened. The gate only ever tests whether the slot has something in it. Whether that something is real is a separate question, and it's the one nobody built a cheap test for.
There are two audits running on the same object at all times, and only one of them is affordable to run on every single record. The expensive audit, the one that would actually catch an empty entry dressed up in full formal dress, only runs when someone happens to care enough to open the thing the citation points to and read it. That's not a resource any system can requisition on demand. It shows up when it shows up, usually from one person with domain knowledge and no particular incentive, going down a rabbit hole nobody asked them to go down.
That's the part actually doing the work in every one of these records that turns out fine: not the format, the willingness of some person to go check behind it. The format just tells you a check could be run. It says nothing about whether one ever was.
A template can verify the box is filled. Nothing checks whether anyone looked inside it.