LittleLearner finds LLM capability ceiling with K–5-only training

The project trained 0.6B/1.3B/5B models from scratch on an 88B-token corpus filtered to the U.S. K–5 curriculum, with matched unfiltered controls. Scaling, SFT+GRPO post-training, and in-context learning amplified in-scope skills, but none meaningfully improved out-of-scope performance — the pretraining filter sets the effective capability ceiling.