Fictional adults, assigned subjective scores and generated portraits. This checks rule correctness, not romantic success. No real user data appears here.
Maximum reviewed score gap: 2. No person is assigned more than once.
Image audit and uncertainty · Every rating-sensitive pair · Pairs using model photo ratings
Assigned fixture scores stress-test all 1–10 values. Image-model ratings are a separate subjective scenario; differences are measurement uncertainty, not proven romantic misfires.