The paper introduces Audit-First VAPO, a method that separates directional admission from magnitude control for selective updates under imperfect verification. It employs an observation-only accept‑appeal‑abstain policy with a finite secondary‑verification budget, freezing the action trace before clean labels are added, and uses finite‑sample bounds to certify harmful risk, coverage, and verifier‑call rate. Experiments on Qwen3.5‑0.8B and GSM8K show that RC‑VAPO achieves high accuracy while maintaining low selected harmful risk and modest verifier cost compared to baseline methods.
arXiv:2606. 08517v1 Announce Type: new Abstract: Selective predictors answer on confident inputs and abstain elsewhere; deploying one safely needs a single finite-sample certificate that simultaneously upper-bounds the selected risk, lower-bounds the acceptance probability $\pacc$ above a floor $\pmin$, and lower-bounds the deployment utility.
arXiv:2609.36958v1 Announce Type: cross Abstract: Repeated verifier calls are useful only when they contribute conditional information. We introduce VStress, an auditable replay contract, and VStress...
arXiv:2607. 29400v1 Announce Type: new Abstract: A routing decision can be revised at the next transaction, but a latched source exclusion persists across later decisions.
The paper demonstrates that a preference‑optimization objective can learn to distinguish reliable from unreliable sources by installing a prior‑dependent reliability switch. By training on data where a source’s stated reliability is paired with its answer, the model learns to flip its response only when the stated reliability exceeds a threshold that grows with the model’s prior. Experiments on Qwen2.5‑7B‑Instruct and Llama‑3.1‑8B show that this switch generalizes to unseen reliability values and follows stated reliability over role prestige, whereas supervised imitation fails to learn it.
arXiv:2609.39123v1 Announce Type: new Abstract: When optimizing an expensive black-box function sequentially, as in hyperparameter optimization, we may want to stop once the best evaluated value is c...
arXiv:2610.18948v1 Announce Type: cross Abstract: Finite safety assessments cannot exclude an adversary that preserves unattacked behavior until certification and activates afterward. We characterize...
arXiv:2610.33112v1 Announce Type: new Abstract: Use-based AI regulation assigns obligations by deployment risk but lacks an evidentiary object to carry them. A benchmark score is a property of a mode...
The paper investigates how verifier components used in post‑training and test‑time loops can collapse when they are refreshed based on aggregate success scores. It shows that as a policy improves on one conjunctive target, the verifier’s on‑policy AUC can shift to favor a worse component, leading to a preference for a full‑outcome predictor over the intended component oracle. The authors provide theoretical thresholds, empirical demonstrations on text control and ECG data, and propose a repair that audits the component on its own failures to prevent damage.
arXiv:2610.09541v1 Announce Type: cross Abstract: Signals that predict whether a chain-of-thought (CoT) trace is correct are compared by AUC, but deploying one requires a threshold with a guarantee....
arXiv:2608. 07914v1 Announce Type: new Abstract: Behavioral contamination detectors can return "no evidence" either because a benchmark is clean or because the audit has little power.
SafeRestore introduces a framework for certifying when an industrial image restoration should be automatically returned to a detector or require human review. It ranks five restoration candidates using action‑specific fitted scores, selects a threshold gate on tuning data, and evaluates the gate on a separate certification sample with two one‑sided exact binomial bounds—one for evidence‑loss incidents and one for excess‑activation incidents. In a retrospective study of 4,591 Carinthia‑S images, the protocol demonstrates auditable risk‑coverage behavior, with varying pass rates across different policies and morphologies.
arXiv:2608. 19760v1 Announce Type: cross Abstract: Audited against causal ground truth from executed replay in a single-agent tool environment (ALFWorld), none of the step-level credit signals used to train LLM agents -- LLM-judge scores, outcome-conditioned logprob ratios, or the policy's own confidence -- identifies which steps causally matter better than chance.