Can we detect category errors?

I just heard this in a YouTube video:
> \- Um, well, world models are a confusing term right now, right?

At first I thought it was an agreement error between plural "models" and "are" vs singular "term".

But that's not it at all since the verb should normally agree with the subject, not the object. (We have a separate issue about flagging that.)

What's really happening is that the speaker/writer is confusing a thing with its name. World models are not terms, confusing or otherwise, they are representations/structures. "world model" is a term.

Corrections would include:
- Um, well, "world model" is a confusing term right now, right?
- Um, well, world models are confusingly named right now, right?
- Um, well, the term "world model" is confusing right now, right?
- Um, well, world models are confusing concepts right now, right?

While suggesting a correction without AI might not be possible, maybe detecting the error would be possible?

The issue names no files, tests, or entry points. Start by determining whether Harper's existing grammar-checking rules can support detecting confusion between a thing and its name without AI. Done means a reliable, scoped detection approach is established for the examples, with its feasibility and test expectations documented.

Written by the indexing model from the issue text.