Why every AI benchmark is quietly grading itself.
The scores look objective. The tests were written by the people being tested.
Placeholder draft. This paragraph stands in for the opening of the long version. It will set out the basic claim: a test tells you something useful only when the people taking it had no hand in writing it.
Who writes the test.
Placeholder draft continues. This section will walk through how evaluations get designed, who tends to design them, and why that matters. The point to land here is simple: a score is a claim about a test, not about the world.
When the test-taker also writes the test, a high score tells you more about the test than about the taker.
Placeholder draft continues. A closing section will cover what a more trustworthy evaluation might look like, and how readers can tell the difference. It will link back to related pieces, such as why your assistant agrees with you too much.