Abstract
Post-training a code generator against a learned judge can optimize proxy features that raise the score without improving the artifact. We study the opposite signal: a deterministic, judge-free, ungameable filter -- whether a generated project launches cleanly under a headless engine (strict-launch).
Under this gate, rejection-sampling self-distillation compounds out-of-family generalization.
On GameCraft-Bench (mapping a natural-language brief to a complete Godot project), a 14B model (Qwen3-14B+LoRA) distilled under strict-launch raises clean generation on four unseen game families from 8.8% to 42.2% per-candidate and best-of-K coverage from 18/25 to 25/25 (the gold ceiling) over three rounds, each a significant gain (p=0.0019).
This method's success indicates that with strict execution gating, generated models can excel not only in known games but also demonstrate robust adaptability when facing entirely new game types.
Blogger's Review: The execution-gated self-distillation approach presented in this paper offers a fresh perspective in game generation, emphasizing real-world performance over mere judge feedback. This innovation could significantly advance the automation process in game development, making it a noteworthy development to watch out for.