The paper examines how pooled leaderboards in microservice root cause analysis (RCA) benchmarks obscure key details about which subsystems the rankings apply to and how scores are generated. By auditing three public benchmark families (OpenRCA, RCAEval, and PetShop) and running 16 released methods on 11 subsystems, the authors show that many reported leads do not hold across all subsystems and that differences in input tables and implementation details can reverse method rankings. They find that none of the 23 RCA papers examined provide the per‑case outputs needed to verify these claims, and they recommend that future pooled comparisons disclose subsystem coverage, score spread, input tables, handling of failures, and denominators, along with per‑case outputs.
arXiv:2606. 29159v1 Announce Type: new Abstract: Offline root-cause-analysis (RCA) benchmarks commonly rank methods by a single pooled top-1 accuracy across multiple subsystems, and engineers often read the pooled winner as a recommendation for their own subsystem.
The paper investigates whether graph structure improves microservice root‑cause analysis by conducting a controlled study on the RCAEval benchmark. Using identical features, optimizers, and evaluation protocols across three model variants, the authors find no consistent advantage for graph‑based models over flat models, with a negligible Avg@5 difference (0.003, p=0.844). They identify two benchmark properties—limited fault injection and a non‑uniform telemetry schema—that bias results, and propose a new model, PSC‑GRCA, which achieves higher Avg@5 mainly through a system prior rather than graph information.
arXiv:2610. 35767v1 Announce Type: new Abstract: Root cause analysis (RCA) of microservice failures has attracted many LLM agents whose investigation is hand-crafted, through standard operating procedures, fixed stages and role agents.
arXiv:2608.22808v2 Announce Type: replace Abstract: When can an agent failure be caught? An audit is usually limited by the record rather than by the method. CatchBench therefore puts one auditor's q...
StrictPRMBench is a replayable evaluation benchmark designed to audit fixed‑score PRM aggregation rules. It standardizes mathematical reasoning pools, scorer evidence, and deletion interventions so that different reducers confront identical selection problems. Across seven checkpoints and two benchmarks, the benchmark reveals that product aggregation yields higher deletion‑induced selection instability than geometric mean, with statistically significant differences in most cells.
arXiv:2607. 01211v1 Announce Type: cross Abstract: Repository-level performance-optimization benchmarks such as GSO, SWE-Perf and SWE-fficiency evaluate coding agents by applying patches to real repositories and comparing runtime against unoptimized baselines and official reference patches.
The paper investigates the reliability of language‑model judges used as measurement instruments on shared endpoints. Through two preregistered audits of 52,988 requests, the authors found that repeat rankings and byte‑identical replays fell far short of required thresholds, revealing significant instability. They identify three mechanisms—label‑to‑meaning bias, candidate gaps below the noise floor, and input permutation noise—that explain the gap, and propose a snapshot‑identity ladder, design rules, and a reporting checklist to mitigate such failures.
The paper reports a preregistered audit of language‑model judges used as measurement instruments, revealing that the assumption that a model’s responses remain stable over time is invalid. Across nearly 53,000 audited requests, repeat rankings and byte‑identical replays fell far below required reliability thresholds, with three identified mechanisms—label‑to‑meaning bias, candidate gaps below the noise floor, and input permutation noise—explaining the discrepancy. The study proposes a three‑level snapshot‑identity framework, eight design rules, and a reporting checklist to prevent such reliability failures in future evaluations.
The study analyzes 373,019 judgments from LLM‑scored benchmarks, decomposing variance into system, item, judge, and interaction components via generalizability theory. It finds that with a single judge, generalizability converges to a ceiling determined by the system‑by‑judge variance, which is substantially lower in pairwise preference settings, allowing one judge to suffice. The research also reveals significant biases in presentation order and highlights that many published win‑rate claims fall below the measured floor of the benchmarks.
arXiv:2609.36686v1 Announce Type: new Abstract: Identifying the root cause of an anomaly among hundreds of sensors is critical for preventing safety incidents and costly downtime in complex monitored...
arXiv:2607. 28545v2 Announce Type: replace-cross Abstract: Large language models can write, patch, and search code, but oncall root cause analysis (RCA) demands something different: reasoning over noisy metrics, logs, traces, and source code, starting from ambiguous user-facing reports, often hours after the incident began.
arXiv:2608. 06940v1 Announce Type: new Abstract: LLM judge panels are a standard evaluation tool, but prior work reports highly correlated panel errors: nine judges provide roughly the effective information of two independent ones, and aggregation closes only a small fraction of the gap.