# Coverage, Not Credit: Failure-Credit Routing of Zeroth-Order Perturbation Budgets Does Not Improve On-Pool Sample Efficiency for LLM Agents

arXiv: https://arxiv.org/abs/2608.28011 · HTML: https://mlsanity.com/paper/2608.28011/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 15 of its 21 references that are in the map.
- First posted on arXiv 28 Aug 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 15 of 15)

- [Evolution Strategies as a Scalable Alternative to Reinforcement Learning](https://mlsanity.com/paper/1703.03864/) (2017)
- [Fine-Tuning Language Models with Just Forward Passes](https://mlsanity.com/paper/2305.17333/) (2023)
- [Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuning](https://mlsanity.com/paper/2402.15751/) (2024)
- [Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning](https://mlsanity.com/paper/2509.24372/) (2025)
- [Evolution Strategies at the Hyperscale](https://mlsanity.com/paper/2511.16652/) (2025)
- [Zeroth-Order Fine-Tuning of LLMs in Random Subspaces](https://mlsanity.com/paper/2410.08989/) (2024)
- [EvoTool: Self-Evolving Tool-Use Policy Optimization in LLM Agents via Blame-Aware Mutation and Diversity-Aware Selection](https://mlsanity.com/paper/2603.04900/) (2026)
- [ESSA: Evolutionary Strategies for Scalable Alignment](https://mlsanity.com/paper/2507.04453/) (2025)
- [EGL-SCA: Structural Credit Assignment for Co-Evolving Instructions and Tools in Graph Reasoning Agents](https://mlsanity.com/paper/2605.10366/) (2026)
- [Learning a Zeroth-Order Optimizer for Fine-Tuning LLMs](https://mlsanity.com/paper/2510.00419/) (2025)
- [Elucidating Subspace Perturbation in Zeroth-Order Optimization: Theory and Practice at Scale](https://mlsanity.com/paper/2501.19099/) (2025)
- [Universally Empowering Zeroth-Order Optimization via Adaptive Layer-wise Sampling](https://mlsanity.com/paper/2604.18264/) (2026)
- [Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs](https://mlsanity.com/paper/2606.05516/) (2026)
- [Cooperative Coevolution for Resource-Constrained Agentic LLM Post-Training](https://mlsanity.com/paper/2608.02391/) (2026)
- [Beyond the Capability Boundary: Zeroth-Order Optimization for Self-Evolving LLM Agents](https://mlsanity.com/paper/2608.09292/) (2026)

## Built on by (most cited first, 1 of 1)

- [Coverage, Not Targeting: A Structural Regime in Multi-Turn Agent Credit Assignment](https://mlsanity.com/paper/2609.02417/) (2026)

## Abstract (arXiv)

Trajectory-level credit assignment can localize which module of a tool-using LLM agent causes failures using only verifiable signals. We ask whether such failure credit should route a fixed zeroth-order/evolution-strategies (ZO/ES) perturbation budget. Across a synthetic environment and frozen Qwen2.5-1.5B/3B and SmolLM2-1.7B agents, three task families, six allocation schemes, a credit-noise sweep, paired seeds, and exact sign-flip tests, we find no statistically detectable improvement over uniform allocation in any on-pool comparison (no gain of at least 2 percentage points). The joint soft-plus-sigma scheme is equivalent to uniform within a +/- 0.02 AUC margin on 1.5B and 3B; concentrating the full budget on the credit argmax is marginally equivalent on 1.5B, where that module is the verified bottleneck, and significantly worse on 3B. Inverse-propensity debiasing does not rescue routing, and misrouting costs up to -0.074 AUC in-house and -0.118 end-to-end on the BFCL-derived family. Across six fixed-step schedules, loss is linear in bottleneck starvation rate (R^2 = 0.94, descriptive), and a preregistered credit-free coverage floor removes detected harm. Matched-budget burst and step-compensating catch-up schedules are consistent with harm arising from insufficient cumulative parameter movement rather than update frequency. Our primary estimand is optimization efficiency on a fixed task pool. On unseen BFCL functions, the study's one exception is that soft routing exceeds uniform on held-out endpoints (+0.047, p = 0.031, n = 6). A plausible but untested reading is that routing-favored caller improvements transfer while uniform's on-pool gains reflect a synthesizer behavior specific to our harness. We report this exception explicitly and document three failure modes that can silently invalidate ZO/ES experiments on frozen LLMs.