Abstract
Imperfect verifiers can assign a harmful update direction even when clipping and regularization bound its magnitude. We introduce Audit-First VAPO, which separates discrete directional admission from continuous magnitude control. An observation-only accept-appeal-abstain policy uses a finite secondary-verification budget; its action trace is frozen before clean labels are joined. Simultaneous finite-sample bounds then certify selected harmful risk, coverage, and verifier-call rate over a predeclared policy family. Conditional Hoeffding-Azuma bounds account for the dependence induced by shared budgets, and rollout or verifier changes initiate a new certification stage. After admission, a bounded trust-clip-KL actuator controls magnitude. We evaluate two models on two reasoning benchmarks against static RLVR, matched-random selection, confidence thresholding, noise correction, and verifier augmentation. On Qwen3.5-0.8B and GSM8K at target risk $ρ=0.08$, RC-VAPO achieves 74.1% accuracy, selected harmful risk 0.0697, coverage 0.4125, and relative verifier cost $1.16\times$. At matched coverage and update magnitude, its selected-risk difference from matched random is -0.0260 with paired 95% interval $[-0.0364,-0.0157]$. Across asymmetric, confidence-dependent, and correlated-verifier noise, the certificate is satisfied on 57 of 60 independent runs. These comparisons isolate informative directional selection from proposal suppression, update shrinkage, and additional verifier computation.
Keywords
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Hu, M., Hu, S., Guo, X., Wang, X., Wang, B., Chen, R., Zha, D., & Xiao, J. (2026). Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification. https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification
MLA 9
Hu, Miaobo, et al. "Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification." https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification.
Chicago (author–date)
Hu, Miaobo, Shuhao Hu, Xiaobo Guo, Xin Wang, Bokun Wang, Rui Chen, Daren Zha, and Jun Xiao. 2026. "Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification." https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification.
Harvard
Hu, M., Hu, S., Guo, X., Wang, X., Wang, B., Chen, R., Zha, D. and Xiao, J. (2026) 'Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification', Available at: https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification.
Vancouver
Hu M, Hu S, Guo X, Wang X, Wang B, Chen R, et al. Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification. https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification
IEEE
M. Hu, S. Hu, X. Guo, X. Wang, B. Wang, R. Chen, D. Zha, and J. Xiao, "Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification," https://omanscience.com/en/articles/audit-first-vapo-risk-certified-selective-updates-under-imperfect-verification.