# The Filter Metric is Safety-Critical: Phantom Advantages in Group-Relative RL under Shaped Rewards

arXiv: https://arxiv.org/abs/2609.13866 · HTML: https://mlsanity.com/paper/2609.13866/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 17 of its 22 references that are in the map.
- First posted on arXiv 12 Sep 2026.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)

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

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [Understanding R1-Zero-Like Training: A Critical Perspective](https://mlsanity.com/paper/2503.20783/) (2025)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Group Sequence Policy Optimization](https://mlsanity.com/paper/2507.18071/) (2025)
- [τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environment](https://mlsanity.com/paper/2506.07982/) (2025)
- [Defining and Characterizing Reward Hacking](https://mlsanity.com/paper/2209.13085/) (2022)
- [Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning](https://mlsanity.com/paper/2508.20751/) (2025)
- [Tackling Length Inflation Without Trade-offs: Group Relative Reward Rescaling for Reinforcement Learning](https://mlsanity.com/paper/2603.10535/) (2026)
- [The Dark Room in the Reward Channel: Dense Prediction Rewards Collapse GRPO-Trained LLM Agents -- and The Channel, Not the Content, Decides What Works](https://mlsanity.com/paper/2607.21273/) (2026)
- [Beyond Penalizing Mistakes: Stabilizing Efficiency Training in Large Reasoning Models via Adaptive Correct-Only Rewards](https://mlsanity.com/paper/2606.22716/) (2026)

## Abstract (arXiv)

Group-relative policy optimization (GRPO and descendants) can discard no-contrast rollout groups through dynamic sampling, while practical implementations expose a configurable filter metric. We identify and quantify a metric-predicate mismatch under composite shaped rewards. When filtering follows the shaped training score rather than the task outcome, all-fail groups retain nonzero within-group spread and pass the predicate; standard-deviation normalization then promotes shaping differences among failures to full-size phantom advantages. In a controlled GSM8K comparison (Qwen2.5-1.5B, LoRA), no filtering and shaped-score filtering end at EM 0.080 +/- 0.112 and 0.040 +/- 0.008, whereas binary-outcome filtering holds 0.754 +/- 0.005 across four runs per arm (three default-seed reruns and one seed-123 run; mean +/- sample SD). On verl's native recipe/dapo trainer, holding model, data, reward and trainer fixed and changing only the metric, the score arm requires no batch refill in any of 40 observed steps and ends at EM 0.160; the accuracy arm refills in 29/40 steps and ends at 0.763. Both use the same custom shaped-reward hook and unmodified trainer/filter code. Prior work established shaping-induced amplification and all-fail filtering; our contribution isolates the metric-predicate semantic mismatch and directly instruments native deletion/refill telemetry. Across tested positive coefficients lambda in {0.1, 0.3, 0.5}, unsafe arms collapse; exploratory one-run cells reproduce the failure at 1.5B/7B on MATH and under GSPO, while disabling standard-deviation normalization avoids the observed collapse. Filtering under a composite reward should use a task-outcome signal whose semantics are independent of shaping.