VeriScale: Adversarial Test-Suite Scaling for Verifiable Code Generation

Yifan Bai, Xiaoyang Liu, Zihao Mou, Guihong Wang, Jian Yu, Shuhan Xie, Yantao Li, Yangyu Zhang, Jingwei Liang, Tao Luo · 2026

Read VeriScale: Adversarial Test-Suite Scaling for Verifiable Code Generation by Yifan Bai, Xiaoyang Liu, Zihao Mou, Guihong Wang, Jian Yu, Shuhan Xie, Yantao Li, Yangyu Zhang, Jingwei Liang, Tao Luo (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-2605-22368.txt.

Description

Bai and colleagues build VeriScale, a tool for making code tests that can be checked by proof. Old test sets are too thin, so models look better than they are. VeriScale grows test sets with hostile code, then trims them to small sets that still tell models apart. On Verina, it makes VerinaPlus, with suites over 83 times as large, and a light VerinaLite at 14 times. Tests on eight top models show VerinaPlus lays bare flaws the old set hid. VerinaLite keeps that power at far less cost.

Catalog record

CREATORYifan Bai, Xiaoyang Liu, Zihao Mou, Guihong Wang, Jian Yu, Shuhan Xie, Yantao Li, Yangyu Zhang, Jingwei Liang, Tao Luo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57620
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.22368

Browse the full library · About the Tower of Babel