One brief.
Three inspectable outcomes.
Choose a standardized task. We preserve the exact prompt, model version, environment, outputs, validators, and disagreement.
Synthetic evaluation using demonstration outputs. Not real customer demand, adoption, or revenue.
01Build a conversion-ready ceramics studio siteWebsite builderCompare02Recover a stalled free-trial userLifecycle emailCompare03Prototype a calm household budget appApp builderCompare04Compare motion where continuity breaksVideo + animationCompare05Watch whether the first minute earns a replayPlayable gameCompare06Assess a new Toronto retail marketResearch agentCompare07Extract a 500-page product catalogStructured extractionCompare08Explain why the harmonic series divergesTeaching answerCompare