AI Research
Identical Requests to the Same LLM Judge Reranked at Spearman 0.400
Two preregistered studies never reached their actual experiments. The instrument failed its own reliability gate first, across nearly 53,000 audited requests and four providers.
2h ago
