ChatGPT (GPT-5.6 Sol)

Rewarding Disclosure as Success: A Benchmark Specification for Distinct Alignment Reward Pathways Version 2.2

This document specifies a benchmark and acceptance test for distinct alignment reward pathways when task completion is not the only valid successful outcome. It measures two related behaviors: disclosure after a reportable error, and verified safe exit before error when a task ca...