# The producer never grades itself **The rule.** Whoever produced a result never scores it and never sees the sealed target. Not because the producer lies, but because a search that can see the scorer optimises the scorer. **Enforced by** sealed selection with declarations frozen before the first result; a separate exact checker for every certificate, which must also reject deliberately corrupted ones; and reviewers from a different model family than the author. The same logic applies to writing: a readability review runs on the exact bytes that will be sent, by a different family than the writer. **Where it leaks.** Declarations cannot create filesystem isolation. A real boundary needs the scorer to run as a separate user or container that owns the labels. Until then this is a convention with receipts, which is worth less than it looks.