Brief fit
Does the output answer the actual audience, channel, constraints, and purpose?
Method v0.1
CreativeBench scores outputs as work products, not magic tricks.
Each run should keep the brief, prompt trail, model/provider, raw output, human edits, final asset, and reviewer notes together. That gives us a receipt, not just a screenshot.
Originality
Does it avoid generic AI mush while staying coherent and usable?
Craft
Is the structure, pacing, wording, visual hierarchy, or concept execution strong?
Revision resilience
Can it improve after critique without collapsing into blandness?
Evidence trail
Are claims, sources, assumptions, and model changes visible enough to audit?
Production readiness
How much human repair is still needed before real-world use?