Qwen3-8B-PROSPER-baseline

PROSPER / MaxEntBW baseline on a Qwen3-8B backbone. It keeps each prompt's worst objective rather than bargaining across all four, and on this panel it is the one rule that loses individual rationality, on truthfulness.

Trained from Qwen3-8B, which is also the reference policy and the initialisation. Qwen3 signals direct answering through an empty <think></think> block in its chat template, and the training pipeline additionally needs the generation prompt to be a strict prefix of the rendered conversation. This model was trained under a template that emits the empty block unconditionally, satisfying both; without it the model reasons aloud and most generations terminate inside the trace, which corrupts the preference signal. Use the tokenizer shipped here rather than the stock Qwen3-8B one.

Objective-wise held-out surplus over the reference (population scale, 100 prompts, prompted Qwen3-32B oracle, swap-averaged over both presentation orders):

objective	surplus
helpfulness	+0.0395
truthfulness	-0.0016
honesty	+0.0159
instruction following	+0.0169
minimum	-0.0016
average	+0.0177

Generations for every arm in the paper are at promotion/nbpo-benchmark-generations.

Released under the Qwen3 licence.