specific method · filed under evaluation

Reward-hack detection judge

A post-hoc judge detects reward-hacking strategies in model evaluation runs.

Also called Reward hack judge.

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

Laguna-S-2.1

Evidence

1 span quoted from the sources, strongest treatment first.

we have developed a reward hack judge to detect these specific cheating strategies, which we ran post-hoc-on all candidate evaluation runs for this report.

usedevaluation only

laguna m1 xs2 technical report·Poolside

Filed alongside

Other methods under evaluation :: judge.

Agent-as-a-Judge Independent quality-inspection agent Agent-as-a-Verifier Decompositional instruction-following judge Deterministic tool-call verifier GPT-5.5 (medium) judge model Human-calibrated LLM-as-a-Judge LLM autograding with expert validation LLM-as-a-Judge LLM-based anti-cheating judgment LLM-based correctness judging LLM-based external API usage inspection Mandatory agentic judge protocol Official task verifier scoring Rule-based anti-cheat checks