promptdojo_

A gate that can't fail isn't a gate — step 5 of 7

This deploy gate has been green for every release in history — including today's candidate, which scores 0.79 against a live model at 0.86. Look at line 5: the threshold is computed from the candidate's own score, so any model beats itself minus 0.02. Green forever, by construction.

Fix line 5 so the threshold comes from the baseline — the score users currently get — and today's regression finally gets caught.

Expected output:

BLOCK
The break is on line 5 — but read the whole snippet first.

full-screen editor opens — close anytime to keep reading.