# itayinbarr.substack.com

Itay Inbarr's Substack at [itayinbarr.substack.com](https://itayinbarr.substack.com). Author of the **little-coder** paper that established the [[scaffold-model-fit]] framing for evaluating small local models on coding-agent benchmarks. Inbarr builds on the SafeRL-Lab/clawspring agent substrate and reports rigorous matched-model experiments — same weights, same benchmark, varied scaffold.

## Ingested articles

- [[little-coder-scaffold-model-fit]] (2026) — Qwen3.5-9B Q4_K_M moves from 19.11% on Aider's default scaffold to 45.56% on little-coder, with mechanism observables for write-tool guards, thinking-budget caps, conditional workspace discovery, and small-budget skill cards.