A 4B coding agent, post-trained only with RL on synthetic software tasks, reaches 61.5% on SWE-bench Verified and beats models many times larger. No distillation from a frontier teacher. The missing ingredient is not a bigger model. It is tasks calibrated to what the current checkpoint can almost, but not quite, solve.

The pipeline estimates how often the agent already resolves a candidate task, then rewrites the ones that are too easy or too hard until they sit in a learnable band. As the policy improves, the curriculum moves: from heavily scaffolded bug fixes toward shorter statements and more diverse change types. A simple tool harness (read, write, edit, glob, bash) lifts the same 4B base far more than it lifts large models. Failures remain mostly reasoning, not tool use. Effective reward hacking is held at zero by the harness, not by hoping the agent is honest.

Directly asking LLMs for numerical ratings (like purchase intent on a 1–5 scale) usually produces unrealistic, overly positive, or narrowly distributed answers that don’t match real human data.

This paper introduces Semantic Similarity Rating (SSR)— a simple method that first lets the LLM generate free-text reasoning about a product, then maps that text to a realistic Likert distribution using embedding similarity to anchor statements.