Askewly Design
Machine checks for what machines can judge. People look at what is left.
The problem
AI output is plausible before it is good. That gap is where slop lives, and it is wide enough that people spend their attention on things a machine could have caught first.
Why better models won’t close it
Taste is a human judgment and stays one. But most of what makes output read as machine-made is mechanical: colors that bypass the tokens, too many type sizes on one screen, contrast that misses. None of that needs an eye.
Where it is now
Building the instrument. The dictionary and the lint pass are live; coverage is still narrow.