mlsanityData to 7 Oct 2026

RouteGuard: Certifying Routing Gain in LLM Multi-Agent Systems When Complementarity Is Not Enough

arXiv 2608.07583 ↗First posted 5 Aug 2026stat.ML cs.LGcited by 1

On the map: LLM agents and tool use → Agent threats, guardrails and governance

Abstract (arXiv)

Multi-agent LLM systems route among model-backed advisors, yet a deployer rarely knows before shipping whether routing will help at all. Prevailing routers optimize a gate's AUC and presume that advisor complementarity suffices. We show that neither determines the deployable gain. We introduce RouteGuard, a deployment-certification framework. Routing gain decomposes as G = π DeltaE, and the achievable gain is governed by a conditional-regret functional Phi, not by AUC. A finite-sample certification bracket comes with a matching Le Cam lower bound, constant-sharp over the fixed-activity class, and a robustness phase transition. On two benchmarks the framework acts as a guardrail. On RouterBench (11 cross-family models) the verdict depends on the sampling unit: the protocol certifies a gain over GPT-4 under prompt-level sampling and withholds it under workload-cluster resampling, because the gain rests on 3 of 86 workload cells. On OpenRCA (three Gemini advisors) the advisors are statistically redundant: the realized oracle sits at or below the independence baseline in all pools we tested (221 RouterBench pools and three OpenRCA distributions), so the protocol correctly refuses to certify. A pre-registered semi-synthetic control confirms calibration: the protocol certifies a genuine gain once m ≥ m^star and does not certify a true null. Code and frozen artifacts will be released with the published version.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use. Builds on 20 of its 31 references that are in the map. First posted on arXiv 5 Aug 2026.

Builds on

20 of 31 references.

References · 20

  1. 2017Selective Classification for Deep Neural Networkscited by 799
  2. 2023FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performancecited by 791
  3. 2025Why Do Multi-Agent LLM Systems Fail?cited by 559
  4. 2020Consistent Estimators for Learning to Defer to an Expertcited by 262
  5. 2017Predict Responsibly: Improving Fairness and Accuracy by Learning to Defercited by 228
  6. 2024RouterBench: A Benchmark for Multi-LLM Routing Systemcited by 222
  7. 2025Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systemscited by 170
  8. 2025Where LLM Agents Fail and How They can Learn From Failurescited by 118
  9. 2025AgenTracer: Who Is Inducing Failure in the LLM Agentic Systems?cited by 105
  10. 2025Universal Model Routing for Efficient LLM Inferencecited by 80
  11. 2025MasRouter: Learning to Route LLMs for Multi-Agent Systemscited by 80
  12. 2022Learning to Defer to Multiple Experts: Consistent Surrogate Losses, Confidence Calibration, and Conformal Ensemblescited by 59
8 more
  1. 2025Causal LLM Routing: End-to-End Regret Minimization from Observational Datacited by 27
  2. 2025Abduct, Act, Predict: Scaffolding Causal Inference for Automated Failure Attribution in Multi-Agent Systemscited by 14
  3. 2026When Routing Collapses: On the Degenerate Convergence of LLM Routerscited by 10
  4. 2025Cer-Eval: Certifiable and Cost-Efficient Evaluation Framework for LLMscited by 9
  5. 2026Why Do AI Agents Systematically Fail at Cloud Root Cause Analysis?cited by 8
  6. 2022Optimal Best Arm Identification in Two-Armed Bandits with a Fixed Budget under a Small Gapcited by 4
  7. 2026Towards Fair and Comprehensive Evaluation of Routers in Collaborative LLM Systemscited by 2
  8. 2026Proactive Routing to Interpretable Surrogates with Distribution-Free Safety Guaranteescited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026What Can Component-Replacement Evidence Establish? A Critical Scoping Review of Local Decisions in LLM Agentscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026ChannelGuard: Safe Models Do Not Compose into Safe Multi-Agent Systems
  2. 2026TwinRouterBench: Fast Static and Live Dynamic Evaluation for Realistic Agentic LLM Routing
  3. 2026The Replay Gap: Static Evaluation of Model Switching in LLM Agents Scores the Wrong World
  4. 2026The Provenance Paradox in Multi-Agent LLM Routing: Delegation Contracts and Attested Identity in LDP
  5. 2026The Same Zero: Why Identical ASR Can Imply Different Guarantees in LLM-Agent Security
  6. 2026Position: A Three-Layer Probabilistic Assume-Guarantee Architecture Is Structurally Required for Safe LLM Agent Deployment
  7. 2026Autonomous Topology Mutation: Safe Runtime Restructuring for Multi-Agent LLM Systems with Capability, State, and Shadow Invariants
  8. 2026PropGuard: Safeguarding LLM-MAS via Propagation-Aware Exploration and Remediation