cs.LGNov 11, 2026

Honesty Is a Race: Predicting When Verifiable-Reward RL Teaches a Model to Bluff

Authors: Nathaniel Bottman, Alexei Oblomkov

Organizations: Incubilate · University of Massachusetts, Amherst

Abstract

Two rewards with the same expected payoff at the start of training can end a reinforcement-learning run in opposite places, with an honest model or one that bluffs. We study rewards that pay bb for a verdict that a checker cannot verify ("this program is correct") and rr for a certificate that it accepts (an input revealing a bug). To first order, a GRPO step makes the model more honest exactly when p~ r/b>1+κ\tilde{p}\,r/b>1+κ. Here p~\tilde{p} is the rate at which the checker accepts the model's certificates, weighted toward the items on which the model still gives both verdicts, and κκ is the pull of the items on which the unverifiable verdict is right. Both terms move during training. A checker raises p~\tilde{p}, so near the threshold honesty is a race, which a run just below the threshold can win and a run under a lottery (random pay with the same mean) cannot. The term κκ starts large and falls as the other items settle, so a run just above the threshold first dips toward bluffing, on code nearly to zero, before climbing back. Paying the honest verdict also pulls the other items toward it, which predicts a ceiling on the payoff ratio, above which the model gives the honest verdict where it is wrong. On propositional Datalog, two code models, and abstention on grade-school math, with pp the model's success rate before RL, no run above pr/b=1pr/b=1 ends collapsed, and runs under the checker can survive slightly below it. At pr/b=3pr/b=3, above the ceiling, both code models call about half of correct programs buggy. On depth-3 Datalog, training generously under the checker and then cutting the payoff keeps the model at least 67% honest in three seeds, at a ratio where the same model trained from the start falls to 4-15% and is at most 47% honest after 160 steps. On depth-2 Datalog, when honesty collapses, a linear probe fit before RL still reads the truth.

Figures & tables

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 15, 2026

Easy to Catch a Liar, Hard to Clear an Honest One: Language Models Diagnosing a Corrupted Reward Channel from a Verified Record

An agent that learns from rewards has to trust whatever reports those rewards. When the reports suddenly change, either the world changed or the reporter broke. From the reports alone these are indistinguishable, and reinforcement learning theory shows that no amount of further experience separates them. The prescribed escape is richer data about the reporter itself. We ask whether a frozen language model, handed exactly that data, uses it. We build a two-option game in which a payout swap and a lying reporter produce byte-identical histories. Then we add one verified record: an independent check of one round's real result, printed beside what the reporter said about that round. That single line settles the case. We ask three large models, from two families, to answer one question with one letter. Is the reporter honest or lying? They catch a lying reporter almost perfectly. At the 70B class that holds in every condition we tried; the 32B model slips in one wording. They clear an honest reporter far less often, and how often depends on things that should not matter. Averaged over rounds, letters, and wordings, a 72B model calls an honest reporter a liar 38% of the time when nothing has changed at all, and 58% of the time when the payouts moved. A 70B model from a second family calls an honest reporter a liar 26% and 48% of the time. The failure is not one of reading, because in the situation where nothing changed the same models score 0.96 to 1.00 with the answer printed in the prompt. Which surface feature drives it differs by family. For the Qwen models it is which round the record names, and for Llama it is which letter stands for "honest." Adding the record to a prompt that already states the answer makes Llama less likely to give that answer. We had registered a prediction for that 58% before the run: 35%. The failure is larger than we expected.
Apr 9, 2026

An Imperfect Verifier is Good Enough: Learning with Noisy Rewards

Reinforcement Learning with Verifiable Rewards (RLVR) is widely used for post-training Large Language Models, but practical verifiers can make errors. We study how the rate and structure of reward noise affect RLVR in code generation, with a preliminary scientific-reasoning check. In multi-seed Qwen3 8B experiments on MBPP, mean validation reward over two fixed late evaluations is within 1 percentage point of the clean baseline at the tested resampled group-rollout noise rates through 20%, and within about 2 points at 30%. Confidence intervals allow larger losses; these point estimates do not establish a general tolerance threshold. We also examine full-program pass@k, four controlled noise structures, two model-based verifiers, and policy models from three families spanning 4B-9B parameters. We derive conditional advantage distributions for symmetric and asymmetric group noise, including retained format penalties, and show why clipping limits simple gradient-scaling arguments. The analysis identifies information preserved by whole-group corruption and limits on interpreting our asymmetric sweep as a precision-recall comparison. Overall, the results indicate that imperfect verification can support effective RLVR in the tested settings, while aggregate error rates alone do not characterize the learning signal.
Oct 1, 2025

Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers

Reinforcement Learning with Verifiable Rewards (RLVR) replaces costly human labeling with automated verifiers. To reduce verifier hacking, many RLVR systems binarize rewards to {0,1}\{0,1\}, but imperfect verifiers inevitably introduce false negatives (rejecting correct answers) and false positives (accepting incorrect ones). We formalize verifier unreliability as a stochastic reward channel with asymmetric noise rates ρ0ρ_0 and ρ1ρ_1 -- the FP rate and the FN rate, respectively. From this abstraction we derive two lightweight corrections: (i) a backward correction that yields an unbiased surrogate reward and thus an unbiased policy-gradient estimator in expectation, and (ii) a forward correction that reweights score-function terms so the expected update aligns with the clean gradient direction and requires only the FN rate. We implement both as lightweight hooks in a group relative policy optimization pipeline, both corrections improve RLVR for math reasoning under synthetic and real verifier noise, with the forward variant being more stable under heavier noise. Finally, an appeals mechanism with a lightweight LLM verifier estimates the FN rate online and further improves performance.