# Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL

arXiv: https://arxiv.org/abs/2605.22217 · HTML: https://mlsanity.com/paper/2605.22217/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 17 of its 20 references that are in the map.
- First posted on arXiv 21 May 2026.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)
- Synthetic data and self-improvement for LLMs / [Self-play RL for reasoning](https://mlsanity.com/synth/self-play-rl-for-reasoning/)

## Builds on (most cited first, 17 of 17)

- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [Absolute Zero: Reinforced Self-play Reasoning with Zero Data](https://mlsanity.com/paper/2505.03335/) (2025)
- [CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution](https://mlsanity.com/paper/2401.03065/) (2024)
- [Can Large Reasoning Models Self-Train?](https://mlsanity.com/paper/2505.21444/) (2025)
- [Beyond Pass@1: Self-Play with Variational Problem Synthesis Sustains RLVR](https://mlsanity.com/paper/2508.14029/) (2025)
- [Goodhart's Law in Reinforcement Learning](https://mlsanity.com/paper/2310.09144/) (2023)
- [Search Self-play: Pushing the Frontier of Agent Capability without Supervision](https://mlsanity.com/paper/2510.18821/) (2025)
- [How Far Can Unsupervised RLVR Scale LLM Training?](https://mlsanity.com/paper/2603.08660/) (2026)
- [SPELL: Self-Play Reinforcement Learning for Evolving Long-Context Language Models](https://mlsanity.com/paper/2509.23863/) (2025)
- [Catastrophic Goodhart: regularizing RLHF with KL divergence does not mitigate heavy-tailed reward misspecification](https://mlsanity.com/paper/2407.14503/) (2024)
- [Self-Play Only Evolves When Self-Synthetic Pipeline Ensures Learnable Information Gain](https://mlsanity.com/paper/2603.02218/) (2026)
- [RLSR: Reinforcement Learning from Self Reward](https://mlsanity.com/paper/2505.08827/) (2025)
- [Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs](https://mlsanity.com/paper/2601.11061/) (2026)
- [An Imperfect Verifier is Good Enough: Learning with Noisy Rewards](https://mlsanity.com/paper/2604.07666/) (2026)
- [Scaling Self-Play with Self-Guidance](https://mlsanity.com/paper/2604.20209/) (2026)
- [From Rebound to Remedy: Understanding and Mitigating Reward Hacking via Representation Engineering](https://mlsanity.com/paper/2604.01476/) (2026)
- [Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning](https://mlsanity.com/paper/2602.21420/) (2026)

## Built on by (most cited first, 4 of 4)

- [Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops](https://mlsanity.com/paper/2607.07663/) (2026)
- [CompassPlay: Rewarding the Proposer for Where It Moves the Solver](https://mlsanity.com/paper/2609.32228/) (2026)
- [Dr. Free: You Don't Need Difficulty Rewards for Self-Evolving Search Agents](https://mlsanity.com/paper/2609.33565/) (2026)
- [When Updating Stops Being Learning: Rethinking LLM Self-Evolution via learnable information gain](https://mlsanity.com/paper/2609.36535/) (2026)

## Abstract (arXiv)

Self-play reinforcement learning trains language models on their own generated tasks, co-evolving a proposer and solver without human labels. Recent systems report strong reasoning gains, but collapse and instability are widely observed and poorly understood. The dominant response treats this as a reward-design problem. We argue instead that self-play stability is governed by two distinct levers: a data-level gate that decides which proposer-generated tasks enter the training pool, and the reward signal that updates the policy on tasks already admitted. Through controlled experiments on a Python output-prediction task and a deterministic-DSL twin task that strips pretraining priors, output ambiguity, and executor noise, we find the two levers are asymmetric. A strict gate is sufficient for stability under every reward variant we test, including a self-consistency reward with no access to ground truth; while no reward variant is sufficient once the gate is removed. This asymmetry exposes a counter-intuitive coupling we call the Grounded Proposer Paradox: a proposer with ground-truth access accelerates collapse faster than an ungrounded one when paired with a self-consistency solver, by concentrating training on clean tasks that form the fastest path to a spurious self-consistent attractor. Replacing the binary gate with a continuous strictness parameter varepsilon further reveals a two-stage phase transition: training-side metrics decouple at low varepsilon, while validation accuracy holds until varepsilon is much higher. Data-level gating, not reward calibration, is the binding constraint on self-play stability.