AI サービスが一時的に利用できないため、復旧後に翻訳を補完します。
The method Every story you read came from the same fixed process. No cherry-picking, no do-overs, no way to tell which lab wrote which story until after you vote. 01 One pipeline Every model gets the same premise, the same prompt series, and the same three-step harness — nothing tuned per model. 02 One run Each story is generated once. What you read is the model's single attempt, not a cherry-picked best-of-many. 03 No retries A rough draft stays in the pool as generated. No re-rolling past a bad output, in either direction. 04 Blind votes Story A and Story B never reveal which model wrote which until after you cast a vote — model identity is not derivable beforehand. Read the full method →