specific method · filed under evaluation
Human-calibrated LLM-as-a-Judge
An LLM judge calibrated with human-labeled trajectories to flag solutions that may reward hack.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
Using an LLM as a Judge (LLMaaJ) calibrated with human labeled trajectories, we tag solutions that potentially reward hack after every eval.
usedevaluation onlyPoolside
Filed alongside
Other methods under evaluation :: judge.
Agent-as-a-JudgeIndependent quality-inspection agentAgent-as-a-VerifierDecompositional instruction-following judgeDeterministic tool-call verifierGPT-5.5 (medium) judge modelLLM autograding with expert validationLLM-as-a-JudgeLLM-based anti-cheating judgmentLLM-based correctness judgingLLM-based external API usage inspectionMandatory agentic judge protocolOfficial task verifier scoringReward-hack detection judgeRule-based anti-cheat checks