The same belief applied to a team's process: who is allowed to approve what, and what "all tests pass" is actually asserting.
GitHub rulesets, editor hooks, and CI gate channels. Ranex asks a separate question: what must the evidence prove?
A session report describes a run. Re-run the command against the artifact on disk before you let that description decide anything.
Stop an agent from approving evidence it produced. The separation-of-duties rule that still matters when models improve.
A single run cannot prove a change helped. Freeze the comparison, set the rules first, and keep observations separate from grades.
A SHA-256 hash identifies approved dependency bytes. It cannot prove that imported code will report your test result honestly.
A citation can be fabricated from memory. Fetch and record the source bytes before you let a design claim borrow their authority.
