cs.LGNov 16, 2026

SpeCal: Spectral Calibration for Unsupervised RLVR

Authors: Kai Tang, Xing Zheng, Haobo Wang, Guangcheng Zhu, Shenzhi Yang, Zhengqing Zang, Yingfan Ma, Cheng Peng, 

Organizations: Zhejiang University · Ant Group

Abstract

Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning abilities of large language models, but its dependence on ground-truth answers or external verifiers limits its ability to exploit abundant unlabeled reasoning data. Unsupervised RLVR alleviates this bottleneck by deriving scalar rewards from the model's own rollouts. However, the reliability of such self-generated supervision varies substantially across rollout groups and individual responses, while existing methods construct or refine their rewards mainly from decoded outcomes and reward statistics. These endpoint signals reveal what answers the model produces, but provide limited information about whether the underlying rollout representations form a reliable structure for learning. Our key observation is that hidden-state spectra offer a complementary view of supervision reliability by characterizing both the organization of a rollout group and the support received by each rollout within that group. Based on this observation, we propose SpeCal, a bi-level spectral calibration framework for unsupervised RLVR. At the group level, SpeCal combines spectral mass with a Gaussian band-pass gate over normalized entropy-based effective rank, favoring rollout groups with substantial representation variation whose effective dimensionality lies in a preferred regime. At the rollout level, the same group effective rank determines how many dominant directions are retained in a leave-one-out spectral subspace formed by the remaining rollouts, within which the conditional support of each held-out rollout is measured. The resulting signals calibrate the original reward-induced advantages at the group and rollout levels without redefining the underlying unlabeled reward.

Figures & tables

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Mar 9, 2026

How Far Can Unsupervised RLVR Scale LLM Training?

Unsupervised reinforcement learning with verifiable rewards (URLVR) offers a pathway to scale LLM training beyond the supervision bottleneck by deriving rewards without ground truth labels. Recent works leverage model intrinsic signals, showing promising early gains, yet their potential and limitations remain unclear. In this work, we revisit URLVR and provide a comprehensive analysis spanning taxonomy, theory and extensive experiments. We first classify URLVR methods into intrinsic versus external based on reward sources, then establish a unified theoretical framework revealing that all intrinsic methods converge toward sharpening the model's initial distribution This sharpening mechanism succeeds when initial confidence aligns with correctness but fails catastrophically when misaligned. Through systematic experiments, we show intrinsic rewards consistently follow a rise-then-fall pattern across methods, with collapse timing determined by model prior rather than engineering choices. Despite these scaling limits, we find intrinsic rewards remain valuable in test-time training on small datasets, and propose Model Collapse Step to measure model prior, serving as a practical indicator for RL trainability. Finally, we explore external reward methods that ground verification in computational asymmetries, showing preliminary evidence they may escape the confidence-correctness ceiling. Our findings chart boundaries for intrinsic URLVR while motivating paths toward scalable alternatives.
Jul 20, 2025

The Invisible Leash: Why RLVR May or May Not Escape Its Origin

Recent advances highlight Reinforcement Learning with Verifiable Rewards (RLVR) as a promising approach to enhancing LLM capabilities. However, it remains unclear whether the current practice of RLVR truly expands models' reasoning boundary or mainly amplifies high-reward outputs that base models already know. We introduce empirical support dynamics, a framework that quantifies how RLVR preserves, expands, and shrinks access to verified solutions. Applied across models, reasoning tasks, and training checkpoints, it matches verified successes question by question to reveal changes that aggregate accuracy can conceal. Across the evaluated model-level benchmark aggregates, the shrinkage of observed support outweighs its expansion: more questions lose a verified success than gain one at the stated sampling budgets. These losses occur alongside gains in pass@1. Some RLVR models increase average token-level entropy while reducing final-answer entropy: their token choices become more uncertain on average even as their final answers become more concentrated. Together, these findings reveal an ``invisible leash'' in current RLVR practice: it can make existing successes more reliable while losing access to other verified solutions. Empirical support dynamics make this imbalance measurable and motivate exploration strategies that reach underrepresented solutions while retaining prior successes.
Apr 28, 2026

JURY-RL: Votes Propose, Proofs Dispose for Label-Free RLVR

Reinforcement learning with verifiable rewards (RLVR) enhances the reasoning of large language models (LLMs), but standard RLVR often depends on human-annotated answers or carefully curated reward specifications. In machine-checkable domains, label-free alternatives such as majority voting or LLM-as-a-judge remove annotation cost but can introduce false positives that destabilize training. We introduce JURY-RL, a label-free RLVR framework that decouples answer proposal from reward disposal: votes from model rollouts propose a candidate answer, and a formal verifier determines whether that candidate can receive positive reward. Concretely, only rollouts matching the plurality-voted answer are rewarded when that answer is successfully verified in Lean. When verification is inconclusive, we invoke ResZero (Residual-Zero), a fallback reward that discards the unverified plurality proposal and redistributes a zero-mean, variance-preserving signal over the residual answers. This design maintains a stable optimization gradient without reinforcing unverifiable consensus. Across three backbone models trained on mathematical data, JURY-RL consistently outperforms other label-free baselines on mathematical reasoning benchmarks and transfers competitively to code generation and general benchmarks. It attains pass@1 performance comparable to supervised ground-truth training, with superior generalization demonstrated by higher pass@k and response diversity.