CRANE : Constrained Reasoning Injection for Code Agents via Nullspace Editing
Mingzhi Zhu, Michele Merler, Raju Pavuluri, Stacy Patterson · 2026
Read CRANE : Constrained Reasoning Injection for Code Agents via Nullspace Editing by Mingzhi Zhu, Michele Merler, Raju Pavuluri, Stacy Patterson (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-14084.txt.
Description
Zhu and three co-authors present CRANE, a training-free edit that injects reasoning into code agents while keeping tool discipline. Paired Instruct and Thinking checkpoints hold complementary but misaligned strengths. Instruct is concise and tool-disciplined, while Thinking plans and recovers better but over-deliberates and drags performance down. CRANE treats the Thinking-minus-Instruct delta as a pool of candidate edits for the Instruct backbone. Magnitude thresholding denoises the delta, and a Conservative Taylor Gate keeps edits that aid both reasoning transfer and tool use.
Catalog record
| CREATOR | Mingzhi Zhu, Michele Merler, Raju Pavuluri, Stacy Patterson |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 130812 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.14084