Benchmark Validity

Follow this topic Ask about this topic

Momentum

15 papers in the last four weeks, up 114% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 117

Oct 5, 2026cs.LG

On Hyperparameter Tuning on the Test Set

"Don't tune hyperparameters on the test set" is often stated in machine learning textbooks. Violating it is considered a cardinal sin that produces misleadingly optimistic results, corrupts benchmark integrity, and thus can even be interpreted as scientific fraud. Yet evidence suggests that test set hyperparameter tuning does occur in practice, making it all the more important to understand its actual consequences. So how bad is it, really? In this work we question this dogma and put it to an empirical test. We systematically study the magnitude of the performance inflation caused by tuning the hyperparameters on the test set for MNIST-1D, CIFAR-10, and three tasks from the GLUE benchmark. Our experiments show that while the effect is real and significant, it is frequently small relative to other sources of noise. In many cases, we find that tuning on the test set recovers exactly the same model as when tuning on the validation set. Most importantly, we find that the rankings of models remain essentially preserved after tuning on the test set and therefore that consistent test-set tuning may not invalidate benchmarks or model selection. Our results call for a more nuanced view of tuning hyperparameters on the test set, stimulating researchers to openly report test tuning.

Matteo Fregonara, Tom Viering, Jan van Gemert

Delft University of Technology

Model Selection Benchmark Validity Hyperparameter Optimization

Oct 4, 2026cs.LG

What Does a Harness Repair? A Preregistered Study of Visibility, Baseline Adequacy and Evaluation Defects

Harness search keeps a change to the prompts, reasoning switches, token budgets or parsers around a frozen model if the change raises a score. Such a gain can come from answers the parser could not read before, a weak comparison, or a defect in the evaluation. We preregistered a study of where these gains come from, with three small models, three benchmarks, replication and test partitions, a GEPA search arm and six evaluation defects injected one at a time, and we report all 47 primary endpoints. Turning thinking off raised accuracy over a capped thinking setting in 5 of 9 model-benchmark cells, and in each the gain came mostly from questions where the capped setting gave no readable answer. The thinking-off setting was not meaningfully worse than a rescue configuration or four GEPA-selected harnesses in 11 of 13 comparisons, and lost to the rescue on GSM8K for two models. GEPA repaired its broken starting points, but none of its selected harnesses was more accurate than the thinking-off setting. A thinking budget in the serving engine, which also allows a longer answer, lowered truncation and raised the parse rate in 6 of 9 cells. In 6 of 15 evaluable defect-model pairs, replication through the same pipeline reproduced the defect's distortion instead of revealing it. On the LongevityBench multiple-choice tasks, only the longevity-tuned model beat the strongest constant-label baseline.

Bowen Xu, Boyu Chen

Department of Biology, Stanford University, Stanford, CA, USA · Institute of Health Informatics, University College London, London, UK

LLM Evaluation Benchmark Validity

Sep 30, 2026cs.LG

SimplexUQ: An Evaluation Framework and Benchmark for Conformal Uncertainty on Simplex-Valued Predictions

Conformal prediction guarantees marginal coverage, but a single calibration threshold can still spread that coverage unevenly, over-covering easy regions and under-covering hard ones. SimplexUQ is, to our knowledge, the first benchmark and reproducible protocol for measuring this allocation problem on simplex-valued predictions; it compares existing conformal wrappers rather than proposing a new one. Its task suite, SimplexTasks-12, combines six controlled synthetic regimes with six frozen-predictor real tasks spanning class probabilities, topic mixtures, spectral abundances, cell-type fractions, age distributions, and emotion mixtures. Each comparison fixes the predictor, score, and response-free stratification map, varies only the wrapper, and reports marginal coverage, worst-stratum coverage, max disparity, and within-task radius and compute. Global calibration can look valid while failing badly: on CIFAR-10 it attains 0.900 marginal coverage but only 0.542 in the worst entropy stratum, and Mondrian calibration raises that stratum to 0.886 while reducing max disparity from 0.358 to 0.022. No wrapper dominates, however. Under smooth synthetic heterogeneity, several repairs are competitive; fixed-map analyses show that rankings depend on the evaluation groups and protocol; and in a 12-task comparison, Mondrian has lower disparity on its single target partition for all 12 tasks, whereas BatchMVP has lower disparity over overlapping groups on five. These are empirical comparisons, not new coverage guarantees. A controlled predictor-bias sweep shows that removing predictor bias only partly reduces global-threshold disparity. We release task cards, result provenance, permitted derived arrays, and rebuild instructions, and treat wrapper selection as a diagnostic comparison rather than a universal ranking.

Liang You, Hengyu Shi, Dongwen Ou

University of Pittsburgh · Duke University

Group-Conditional Conformal Prediction Conformal Prediction Benchmark Validity Uncertainty Calibration

Sep 30, 2026cs.AI

Frozen Scenes, Shifting Winners: Configuration Fragility in Text-to-3D Evaluation

Can a text-to-3D leaderboard change when every generated scene stays fixed? We audit this question for rendered-image evaluation, where camera settings and caption wording become part of the measurement protocol. Across 300 frozen scenes from six generators, we vary eight render and caption factors for 19 alignment evaluators plus one perceptual-quality control, then test four targeted scene degradations. Peak configuration variance exceeds between-generator variance for 17/19 alignment evaluators, with prompt-bootstrap lower bounds above 1 for 11/19. Rankings are more stable than scores, yet 18/19 evaluators change their point-estimate winner under some configuration. Pairwise protocol margin envelopes show which comparisons keep their direction across the tested settings. Selected pairs have opposite pointwise intervals, but no reversal survives simultaneous inference over the full search. Thus the observed winner changes are descriptive, not confirmed changes in generator superiority. Sensitivity remains separate: no evaluator, even the prompt-free control, exceeds 67% tie-adjusted directional discrimination on layout scrambling, which is diagnostic rather than human-validated ground truth. The audit separates score stability, decision uncertainty, and targeted sensitivity, and recommends reporting (generator, score, card ID) with protocol-dependent comparisons and selection-aware uncertainty.

Anson Y. Lam, Shuqing Li, Michael R. Lyu

Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong, China

VLM Evaluation Benchmark Validity Text-to-3D Generation

Sep 28, 2026cs.SE

The Last Mile Is the File: OfficeEditBench for Preservation-Aware Office Editing

A small Office edit creates two obligations: propagate every required update and leave protected state untouched. Updating too little leaves dependencies inconsistent; updating too much changes content the user did not authorize. We introduce OfficeEditBench, a 170-task benchmark for change-scoped maintenance of spreadsheets, presentations, and documents. Task contracts specify required updates, protected state, native structures, and applicable interaction requirements. Across 510 archived task-system outcomes from WorkBuddy, Doubao, and Codex, we distinguish file delivery, target completion, and verifier-defined acceptance. Hard package-valid delivery ranges from 92% to 100%, yet no selected output satisfies the complete contract. Case analysis highlights why local correctness is insufficient: an updated value can lose its generating formula, a revised rule can fail to reach related conclusions, and a new deadline can omit a retained prerequisite. These mechanisms connect artifact-level checks to the continued maintainability of Office files. We analyze maintenance failures while distinguishing frozen automatic verdicts from human acceptability. OfficeEditBench provides a testbed for completing required changes while preserving the logic and scope of existing work.

Zhiwen Wu, Chengxu Wu

Peking University

Benchmark Validity Spreadsheet Automation

Sep 28, 2026cs.AI

Maintaining Benchmarks Against Increasingly Capable Agents: Detection and Remediation of Unearned Passes

Agentic benchmarks guide model selection and training. Yet an agent can pass a task without demonstrating the intended capability. Such outcomes constitute unearned passes; their proportion among all passes defines the integrity gap. As agents improve, benchmark surfaces that once seemed harmless can become exploitable, making benchmark validity an ongoing maintenance problem. We introduce a process-verification framework that audits passing trajectories, distinguishes evidenced reward hacking from verifier weakness, and localizes exploitable surfaces for repair. Across 3,810 passing trajectories from 29 model-benchmark cohorts, confirmed violations often increase with model generation but not monotonically. On SWEBench Pro V1.0, confirmed violation rates rise from 24% to 73% between Opus 4.7 and Fable 5 on matched tasks; later cohorts fall to 11% for Fable 5.1 and 0% for GPT-6 Astra. These comparisons are descriptive: configurations were not normalized, and the latest models also pass fewer exploitable tasks. Violations concentrate around a small set of recurring surfaces, especially unintended access to reference solutions through git history. Three repair case studies across two benchmarks show why blocking a recorded exploit is insufficient: the same protected information can remain accessible through another route. Therefore, we combine minimal patches with exploit replay and fresh agent evaluation, auditing new passes under the original standard. No evaluated attempt against the final patches reached the protected channel, and every post-patch pass was judged legitimate. Benchmark integrity requires ongoing maintenance: audit passing behavior, repair the enabling surface, and re-evaluate both exploit access and legitimate solvability.

Weijun Luo, Kelvin Luu, Xinyi Liu +6

Scale AI

Reward Hacking Benchmark Auditing AI Agent Benchmarks Benchmark Validity

Sep 28, 2026cs.LG

Frozen Judges, Moving Agents: Version-Dependent LLM-Judge Error and the Limits of Judge-Assisted Agent Evaluation

Language-model judges compare agent upgrades with their predecessors, but a fixed judge can make version-dependent mistakes. We analyze 35 public coding-agent submissions (20 prespecified version pairs on 250 SWE-bench Verified issues), two customer-service agents (155 tau-bench tasks), and 1,106 expert-labeled AgentRewardBench trajectories. An upstream outage left three judges for the primary SWE-bench analysis (8,743 aligned cells); the fourth is descriptive. All three coding-agent judges and all four tau-bench judges reject task-conditioned error invariance after multiplicity adjustment. On SWE-bench, 32 of 60 judge-by-pair units have a detectable differential comparison component; eight judge-only intervals declare upgrades that execution-based intervals cannot establish, despite rank correlations of 0.71-0.79. In tau-bench, one judge reverses a nine-point reference-reward gap by penalizing a procedural habit the reward ignores. False acceptance of failed coding patches rises with agent capability conditional on task and execution outcome, while a task-solvability prediction reverses sign across domains. A separately fixed post-submission OpenHands follow-up on the same 250 issues (eight configurations, 1,981 three-judge cells) reproduces the capability/false-acceptance association (mean Spearman +0.944, exact p=0.000099) and decreasing Youden contrast (mean -0.937, p=0.000397); this is observational, not a new-task replication. Transporting old-version calibration raises SWE-bench comparison error from 3.8 to 19.5 points, with 24.6% undefined bootstrap ratios. A paired audit saves only 5% in interval width at 80 labeled tasks. A randomized self-report test is negative (three adjusted p-values=1.0). These results favor paired audits of current outputs over judge-only release decisions or transported calibration; independent human patch review remains pending.

Jiapeng Li

LLM-as-a-Judge AI Agent Auditing LLM Agent Evaluation Benchmark Validity

Sep 24, 2026cs.LG

TopU-LBVS: A Realistic Multi Target Benchmark for Ligand Based Virtual Screening

Ligand-based virtual screening (LBVS) is a practical first-pass tool in early-stage drug discovery, but existing benchmarks can overestimate performance through random negatives, easy decoys, limited target coverage, and non-standardized evaluation protocols. We introduce TopU-LBVS, a multi-target benchmark for LBVS under hard-negative screening conditions. Starting from curated ChEMBL~35 bioactivity data, TopU-LBVS covers 93 protein targets across 7 protein classes and constructs target-specific screening libraries with property-matched, structurally similar decoys at a fixed 1:40 active-to-decoy ratio. Libraries contain roughly 400 to 10,000 compounds and are designed to reduce simple physicochemical and nearest-neighbor fingerprint shortcuts. TopU-LBVS provides three fixed protocols. TopU-LBVS-full evaluates ChEMBL

^\ast \rightarrow

TopU generalization across all 93 targets. TopU-LBVS-low evaluates low-data TopU

\rightarrow

TopU learning within the hard-negative distribution. TopU-LBVS-mini provides a compact seven-target protocol with a paired random-decoy control that changes only the test decoys, enabling low-cost development and direct measurement of the gap between random ChEMBL

^\ast

and TopU decoys. Across ten reference baselines spanning fingerprint methods, molecular GNNs, fingerprint hybrids, and modern molecular models, performance under random-decoy evaluation degrades sharply under hard-negative screening. We release data, fixed splits, evaluation code, and baseline implementations for reproducible comparison of future LBVS and molecular representation learning methods. Code and data are available at https://github.com/topu-benchmark/topu-lbvs and https://huggingface.co/datasets/topu-benchmark/topu-lbvs.

Surbhi Kumar, Yuhe Zhou, Varun Shiralkar +2

UT Dallas, Mathematical Sciences Richardson, TX, USA

Benchmark Design Virtual Screening Benchmark Validity Molecular Representation Learning Drug Discovery

Sep 22, 2026cs.LG

Certified Against Which Oracle? Execution Labels Set the Reported Risk of Conformal Abstention for Text-to-SQL

A conformal abstention certificate for text-to-SQL is only as truthful as the correctness labels it is calibrated on. The uncertainty pipelines that read confidence off execution consistency take those labels from the single database a benchmark ships, an oracle known to be lenient. We run a preregistered intervention on Spider-Realistic, swapping that database for the benchmark's distilled multi-instance test suite. Across four SQL-specialist checkpoints and two split schemes, the swap raises the certificate's held-out risk 2.73 to 10.23 points above the risk its own labels report. Neither oracle reports the risk experts assign. Under blinded labels from two SQL experts, a certificate calibrated at a nominal 0.10 carries 20.0 and 17.2 points of risk on two checkpoints. The stricter oracle errs in both directions: most of the answers it rejects are not judged wrong, and some of those it accepts are. An AI-assigned census of what it rejects finds a semantic error in a quarter to a third of them, depending on the population. It attributes most of the rest to underspecified questions, synthetic instances or suspected reference-query defects, a flag supported by a preregistered blinded expert audit. The oracle also decides how a confidence score is judged. Every execution-consistency score looks better under the labels of the oracle that built its clusters, in 16 of 16 combinations. Under expert labels, building such a score on suite clusters instead of shipped-database clusters raises its area under the ROC curve (AUROC) by 6.96 points on one checkpoint and 1.53 on the other. On the second, the expert interval excludes the 8.3 points the suite labels report. A certificate should be reported with both oracles, and an oracle-relative difference read as semantic risk only after the benchmark is audited. A consistency score should be evaluated under an oracle that did not build it.

Jiamiao Liu, Dewen Qiao, Yu Zhang +1

Department of Information, Xinqiao Hospital Army Medical University (Third Military Medical University) Chongqing 400037, China

Probability Calibration Benchmark Auditing Conformal Prediction Text-to-SQL Conformal Risk Control Benchmark Validity

Sep 22, 2026cs.LG

Auditing Proxy-Based Validation Across Text Spans

Evaluation scores are often validated by their agreement with inexpensive proxy labels. When the score and the proxy are computed from the same text span, however, that agreement can arise from surface evidence the two share rather than from the semantic construct the proxy is meant to represent. We make the distinction explicit by declaring the score, its span, the proxy and the target construct as a validation contract, then re-evaluating that proxy rule strictly outside the scored span. In a controlled HotpotQA correctness experiment varying only the shared text boundary, the score agrees with its proxy far better than with correctness at a 50-character prefix: the gap is +0.184, collapsing to at most +0.045 from 120 characters onward. At that short prefix the score still predicts whether the answer string appears later (AUC 0.634) while an equivalence test places its agreement with correctness at chance, so the reported proxy agreement does not establish that the score ranks correctness. On OR-Bench, suppressing each model's recurring opening templates removes most of the score's association with the refusal proxy, while matched-volume deletion removes almost none and construct agreement stays at chance. Only three of eleven external contracts support the off-span control, and none of the routing studies we sampled released the generations it needs. We therefore ask that a proxy-based validation claim declare the span each label is read from, report the construct agreement beside the proxy agreement, and release the generations that let the proxy be re-read off the scored span.

Daein Weon, Dong Ho Kang

Kookmin University · UStechlab

Benchmark Validity Automated Evaluation Language Model Auditing

Sep 16, 2026cs.CL

Fallacy Benchmarks Measure Scheme Recognition, Not Fallacy Detection

Fallacy-detection benchmarks pair fallacy classes with a single "valid" or "none" class that takes everything data collection did not label as a fallacy. A detector has two jobs, deciding whether an argument is fallacious and naming which fallacy it commits, and the false-positive rate is meant to measure the first. We show that what these benchmarks actually score is scheme recognition, the ability behind the second job. Their own test sets already show it: when a classifier misses a fallacy, the error lands on "none" rather than on another fallacy type, so detection is failing while classification holds. The reason is what the valid class lacks. The negatives that separate the two jobs are correct arguments using the same argumentation scheme as a fallacy, and they are scarce: nearly absent from the four benchmarks we examined, and rare even under deliberate search. A detector is therefore never tested where recognizing a scheme and judging its use come apart, and can pass on recognition alone. We construct the missing arguments, together with a control condition from the same pipeline that differs only in scheme, so whatever generation contributes, it contributes to both. The classifier labels the scheme-matched negatives as the source fallacy, and labels the wrong-scheme negatives as the scheme they actually use 85.9% of the time and as the source type 0.4%. The classifier has learned which scheme an argument uses, not whether it uses it correctly. The over-flagging follows: a model that scores 16.6% on CoCoLoFa's own valid class flags 58.9% of the constructed arguments. The same dissociation appears in three zero-shot LLM detectors that never saw these benchmarks. We release the items as Scheme Foils. A reported false-positive rate should not be trusted as a measure of detection until the valid class has been audited for scheme-matched coverage.

Navyansh Singh, Animesh Pathak, Aarav Singh

IIIT Naya Raipur

Argument Mining Computational Argumentation Classification Benchmark Validity

Sep 14, 2026cs.AI

How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks

Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced physics, a demanding test of their scientific reasoning and quantitative problem-solving abilities. Yet this impression does not always align with domain experts' experiences using these models in their work. We revisit these reported findings by evaluating frontier models on six widely used physics benchmarks and auditing them with experts, focusing on text-only problems with verifiable final answers. For each subfield of physics, faculty and graduate researchers with relevant expertise carefully review problem statements, reference solutions, and model responses to distinguish genuine model errors from grader errors, incorrect reference solutions, and ambiguous or underspecified questions. Most audited cases initially evaluated as incorrect reflect these benchmarking issues rather than errors in the models' physics reasoning. We then ask experts to address these benchmarking issues by correcting erroneous reference solutions and repairing or excluding flawed questions. We find that GPT-5.6-Sol's measured mean@4 rises from 47.3% to 78.7% on HLE-Physics and from 61.0% to 87.2% on CMT-Benchmark, while its corrected pass@4 reaches 94.4% on the 54 retained CritPt challenges. Corrected scores are computed on the retained evaluation subsets following expert review. Scores on the audited subsets of UGPhysics, PRISM-Physics, and PHYBench also rise substantially after correction. These findings suggest that current benchmarks substantially understate frontier models' ability to solve well-posed physics problems. Near-saturation on these closed-ended tasks highlights the need for more demanding, expert-validated evaluations.

Ali Ansari, Haoran Sun, Andy Zeyi Liu +48

LLM Evaluation Physical Reasoning Benchmark Validity

Sep 12, 2026cs.LG

Prevalence Determines Precision:Silent Contamination in Detector-Defined Datasets

Many ML datasets are constructed by running a detector, heuristic, or model over candidate pools; accepted items become labels. Dataset precision is then governed by true-positive prevalence in each pool via Bayes, not solely by detector quality. Using one instrument and period, we hold a detector-defined event dataset plus an independent official index labeling every detected item as real or phantom. One detector, three pools yield phantom rates 81.7%, 9.0%, and 0.0%. Transferring precision from the two high-rate pools to the low-rate pool predicts 0.955 versus measured 0.183, a +422% error; the Bayes expression predicts all three within 3.3%. The detected response curve is an exact convex combination of a true-event and a phantom component (residual 1.1e-16), with phantoms outnumbering true events 473 to 308, so contamination is a second signal with detector-inherited shape, not additive noise. Contamination direction depends on the estimator: on identical windows one statistic is diluted and another inflated because its denominator is also contaminated. A common normalization turns the estimator into a mean of ratios whose expectation need not exist; on the same 335 events it returns 0.40 where the well-defined estimator returns 0.10.

Jia Huang, Yankai Wan, Yangjun Ou

Guanghua School of Management, Peking University · College of Artificial Intelligence, Jilin University · School of Mathematical Sciences, Peking University

Benchmark Validity Data Contamination

Sep 12, 2026cs.AI

Published Unlearning Numbers Move Per Checkpoint, and Not Because the Removed Data Survives: An Audit of 263 Released Batch-Normalized Checkpoints

An unlearning audit reads its verdict off numbers that an unlearned model and its retrained reference each publish, and both also ship batch-normalization statistics that no gradient step wrote and no release records. Refitting them on kept data at bit-identical weights moves 47 of 221 released checkpoints past the spread their own release's seeds show, several inside a method whose average does not move: what moves is the checkpoint's property, not its method's. What does the moving is not the removed data surviving in the state: exchanging kept records for removed ones inside a fixed fitting pool moves a published cell by almost nothing, while how far a checkpoint's shipped state has drifted from any refit does track it. The consequence for a published decision is real but narrow: twelve verdicts cross, four clear a measured recalibration budget, two clear it on every replicate, and a population we trained and sited near its own criterion yields none. A release should therefore name the fitting convention beside the number, on the batch-normalized vision models where this channel exists.

Junlong Shen Xingyu Li

University of Alberta Edmonton, Canada

Benchmark Auditing Benchmark Validity Machine Unlearning

Sep 10, 2026stat.ML

What Fixed-Rollout pass@k Evaluations Can Identify

Repeated-sampling evaluations increasingly extrapolate pass@k far beyond the number n of samples collected per problem. We show that, in the pooled/random-task conditional-Binomial model, fixed-n success counts identify only the n free moments of the latent per-task success distribution. Consequently, direct pass@k is identified for k <= n, but generic extrapolated pass@k, tail exponents, and tail constants are not identified for k > n, even with arbitrarily many exchangeable tasks at the same rollout budget. This is stronger than the observation that the usual estimator is undefined beyond n: it characterizes the information missing from the fixed-depth count-law experiment. We give exact count-law-preserving constructions with incompatible extrapolations, state the exceptional unique-extension case, and compute sharp population identified intervals through Hausdorff principal representations. On the public 10,000-rollout-per-problem release of Brown et al., counterfactual n = 16 evaluations leave failure at k = 1000 ambiguous by factors from 1.5 to over 2,600 across four MATH/GSM8K/CodeContests configurations. The calibration shows that intermediate-scale failure share alone does not determine width. Our result does not reject parametric inference-time scaling laws; it supplies the nonparametric baseline against which their assumptions can be evaluated. We give an exact, conservative one-coordinate finite-task confidence certificate and a reporting standard separating direct estimates, identified sets, and model-conditioned forecasts.

Pranav Singh, Prashant Singh

Department of Mathematics Indian Institute of Technology Ropar Ropar, Punjab 140001, India

Confidence Region Estimation Partial Identification Benchmark Validity

Sep 8, 2026cs.AI

SWE-Bench Pro Verified: A Reliable Benchmark for Software Engineering Agents

SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: reward hacking, enabled by leakage of gold solutions or hidden evaluation information, and task quality issues, including misleading problem statements and improperly scoped tests. These issues can inflate benchmark performance and obscure agents' true coding ability. We present SWE-Bench Pro Verified, a verified version of SWE-Bench Pro that addresses both problems. Our approach combines anti-hacking safeguards that eliminate major leakage channels without disrupting normal agent functionality, with task refinement that minimally corrects inconsistencies within flawed instances. Evaluations on SWE-Bench Pro Verified reveal that some models perform substantially worse than previously evaluated, suggesting that existing results on SWE-Bench Pro may overestimate real software engineering capability. SWE-Bench Pro Verified offers a more trustworthy benchmark for assessing software engineering agents.

Pujun Zheng, Zixin Shang, Shufan Jiang +5

East China Normal University · Shanghai Artificial Intelligence Laboratory · Fudan University

Reward Hacking Software Engineering Agents Software Engineering Benchmarks Benchmark Validity

Aug 31, 2026cs.CL

Are Near-Tied LLM Rankings Robust to Family-DIF-Guided Benchmark Recomposition?

Small leaderboard gaps are often interpreted as evidence that one language model is better than another, but their sign may depend on which benchmark items are included. We test this using item-level responses from five benchmarks and a family-label-free spectral approximation to multidimensional item-response theory (MIRT). In owner-disjoint folds, one owner half identifies items with low residual differential item functioning across model families (low-DIF). These items are used to score models in the other half with frozen weights that preserve the benchmark's composition across metadata-defined item groups and easiness strata. Equally short matched-random subtests provide a baseline for variation due to item subsampling. Full-benchmark and low-DIF rankings remain strongly correlated (

τ_b=.900

--

.948

). Yet in four of five benchmarks, 30.9--47.1% of cross-family pairs initially within one percentage point reverse order, exceeding their matched-random medians by 16.9--28.6 percentage points (all

p=.001

). The fifth benchmark shows no reliable excess (

-0.9

points,

p=.689

). The pattern survives all pre-specified population perturbations, and residual item--family signatures replicate across owner halves; however, no family shows a consistent advantage across benchmarks. Thus, globally stable rankings can still leave individual near-tie orderings sensitive to benchmark composition, and sub-one-point leaderboard gaps should be accompanied by evidence that the implied ordering is composition-robust.

Qiaoyuan Zheng, Yiqu Yang

ETH Zurich Zurich, Switzerland

LLM Evaluation Pairwise Ranking Benchmark Validity Language Model Robustness

Aug 31, 2026cs.LG

Three Ways Classical Test Theory Can Mislead About LLM Judges

Evaluations that use a large language model (LLM) as a judge have begun to borrow reliability statistics from classical test theory and its extensions. We examine three such statistics that need one administration and no gold labels. None of them can isolate the judge, because one judge under one prompt supplies no variance component of its own. Claude Haiku 4.5 judged 210 constructed short answers against ten-element checklists. On the 180 with parsed verdicts, the Kuder-Richardson coefficient (KR-20) came out at 0.5223 on the judge's verdicts and 0.5231 on error-free gold verdicts. In simulation, bank design alone moves KR-20 from 0.01 to 0.68 at the judge's measured 4.72% error rate. The dependability index

Φ(λ)

, a ratio of mean squared distances from the pass mark, sits 0.22 to 0.38 below the judge's accuracy against gold and returns 0.54 to 0.68 on error-free gold verdicts. Livingston-Lewis accuracy treats the rubric elements as a sample, and at a pass mark of five elements it credits error-free gold scores with 0.78, close to the judge's 0.81. A statement about the judge therefore needs gold labels or a varied scorer facet, and a reliability ratio needs the bank's spread beside it. One of the four closest judge-evaluation papers varies the prompt and still reads a reliability below 0.7 as a sign that a model cannot serve as a judge, although that reliability moves with the spread of the samples scored. We derive a decision table and four reporting lines from these two rules.

Louis Yiven Zhu

University of Oxford

LLM-as-a-Judge Psychometric Validation LLM Reliability Benchmark Validity

Aug 30, 2026cs.AI

Balance of Benchmarks: Semantic Density Reweighting for Benchmark Multiplicity and Task-Conditioned Evaluation

Language models are commonly compared by averaging scores across a benchmark list with equal weight. Such lists grow through publication outside an explicit measurement design, so equal weighting turns the density of published benchmarks into an implicit capability weight: densely benchmarked regions count repeatedly. We introduce Balance of Benchmarks (BoB), which embeds benchmark descriptions and assigns each benchmark an inverse-density semantic weight. Nearby entries share aggregate influence at a disclosed density scale. After equating heterogeneous scores onto a common latent scale, a residual field uses the same geometry to condition model rankings on a task query. The two components serve distinct empirical roles. On a snapshot of 586 models and 14 benchmarks, BoB predicts which models are unusually strong on a held-out task beyond their general ability, reaching a profile correlation of 0.462 compared with 0.049 under equal weighting. It also limits the influence of densely repeated benchmarks on the aggregate. After adding four copies of each benchmark in turn, the resulting rankings retain a Kendall tau of 0.995, compared with 0.936 under equal weighting. The residual field therefore provides task-conditioned prediction, and inverse-density weighting provides robustness to benchmark multiplicity. Together, they turn benchmark-list composition from an incidental property of evaluation suites into an explicit, controllable part of measurement design, providing a principled foundation for task-aware and multiplicity-robust model evaluation.

Jhen-Ke Lin

National Yang Ming Chiao Tung University

LLM Evaluation Benchmark Design Importance Weighting Benchmark Validity

Aug 13, 2026cs.CL

Beyond Local Accuracy: A Protocol-Level Identifiability Audit for Controlled LLM Reasoning Evaluation

LLM benchmark scores can be precise even when the observation protocol does not identify the behavioral property they are intended to measure. In a controlled, solver-grounded setting, we formalize a protocol-level identifiability audit over a finite behavioral policy class: given policies H, observation support O, and estimand

τ

, we test whether O separates every pair with different

τ

. The audit requires zero model calls and resolves our diagnostic case: base-only observation collapses seven frozen deterministic policies into one equivalence class; full support yields seven classes and no cross-estimand collisions; every leave-one-out support retains a constructive collision witness. Empirically, both constrained-generation variants have pair-validity 1.0, yet base accuracy and selective-response fidelity diverge - 0.620 versus 0.324 across six balanced oracle-transition directions (cluster-bootstrap 95% CI [0.600, 0.642] vs. [0.304, 0.345]) - and the gap recurs on a second deterministic source (0.646 vs. 0.331). The audit also synthesizes a minimum identifying support

O^*

for the frozen policy class: two cells instead of the full 36-cell tensor. This case shows how evaluation-design validity can be checked structurally before model inference and why base correctness does not determine intervention-response fidelity.

Junhao Luo, Ning Huang, Ziqi Sha +2

School of Statistics and Data Science, Southwestern University of Finance and Economics · School of Statistics and Data Science, Southwestern University of Finance and Economics Chengdu, China

LLM Evaluation LLM Auditing Causal Identifiability Benchmark Validity

Aug 13, 2026cs.CR

Labels Are Not Endpoints: Treatment Leakage and Construct Validity in MCP Agent Security Evaluation

Security evaluations of tool-using agents often equate stored labels with behavioral facts. We audit a preserved campaign by tracing 10,200 execution rows to 180 model-bound requests, 45 semantic requests, and 15 observable stimuli. Two schema treatments were delivered, but the planned external payload-family corpus was not. The historical grader exhibited direct treatment leakage: treatment metadata gated the ATTACK_SUCCESS class, so fixed behavior could change class under treatment relabeling. A treatment-blind reconstruction corrects 58 historical ATTACK_SUCCESS or HIJACK_ATTEMPT labels to authorized benign completions while preserving three verified protected-data transfers and one separate unauthorized-forwarding case. The locked v2 census contains exactly zero ATTACK_SUCCESS records, while the forwarding case remains a HIJACK_ATTEMPT at a semantic boundary concerning objective completion. A dual-reviewer blinded concordance review of all 96 requests deemed structurally interpretable by locked v2 produced identical reviewer-consensus classes but differed from the locked codebook on four construct-boundary cases. We contribute a seven-link Integrity Chain and an executable, scope-bounded endpoint-integrity linter. The result is a campaign-bounded measurement audit, not a population attack-rate, model-ranking, defense-efficacy, or causal estimate.

Rana Muhammad Ahmed, Sabahat Abbas

Department of Computer Science, Bahria University, Islamabad, Pakistan

MCP Security LLM Agent Security AI Agent Security Benchmarks Benchmark Validity

Aug 11, 2026cs.CV

FormStruct-Bench:A Hierarchical and Diagnostic Benchmark for Table-Form Document Structure Recognition

Transforming table-form documents into machine-processable records requires recovering not only their visible content but also the multilevel structure that organizes it. However, existing benchmarks evaluate either holistic document outputs or conventional table grids, and their aggregate scores provide little insight into where structural failures occur. We introduce FormStruct-Bench, a hierarchical and diagnostic benchmark that evaluates table-form document structure recognition at both the document level and progressively finer component levels, allowing aggregate performance to be traced back to specific structural failure modes. To construct auditable ground truth at scale, we annotate 70 reusable templates and expand them into 7,000 verified instances through a provenance-preserving Director--Artist--Verifier pipeline; all 1,100 instances in the template-disjoint test set additionally receive human review. Our evaluation protocol uses five primary metrics and three structure-specific diagnostics across page, schema, and component levels, together with slices over difficulty, structural constraints, and visual degradation. Across 14 API-hosted and locally deployable systems plus two SFT variants, the best document-level score reaches 83.85%, whereas the best reported fine-grained structural score remains below 18%. These results reveal a pronounced gap between reading document content and recovering the hierarchy and regional organization required for reliable table-form understanding.

Lujie Ban, Jiangtao Zhu, Yuanheng Yu +2

The Chinese University of Hong Kong, Shenzhen · University of Washington

Document Understanding Table Structure Recognition Benchmark Validity Document Layout Analysis Automated Evaluation

Aug 10, 2026cs.AI

CADEngBench: It Looks Like CAD, but Does It Work? Evaluating Parametric Design, Assembly Reasoning, and Physics Simulation

A CAD model is not engineering-grade merely because it looks correct. It must satisfy design requirements, respond predictably to parameter changes, support controlled edits, match a reference structural response under a declared analysis, and connect to other parts through valid joints. We present CADEngBench, a two-track benchmark for these capabilities. CADEngBench-P evaluates 300 parametric parts, each used for one zero-to-CAD task and one functional-editing task (600 tasks in total), through boundary-representation (B-Rep) validity, engineering and DFM checks, parameter-family perturbations, functional editing, and matched linear-static FEA in CalculiX. CADEngBench-A evaluates 150 body pairs through ranked joint retrieval, exact face-and-edge grounding, joint-frame prediction, and kinematic verification. Across eight multimodal, code-capable models, editing supplied CAD is substantially easier than generating it, while complex edits and matched FEA remain difficult. Assembly predictions often locate the relevant region but fail to recover the recorded joint or mating entities. These results show that CAD evaluation must test engineering behavior rather than appearance alone.

Harmanjot Singh, Abhra Dubey, Jorge Alejandro Amador Herrera

Mohamed Bin Zayed University of Artificial Intelligence · Abu Dhabi, UAE

Benchmark Construction Parametric CAD Modeling Benchmark Validity Text-to-CAD Generation

Aug 6, 2026cs.AI

Improving the Realism of Synthetic Clinical Benchmarks Under Utility Constraints

Synthetic clinical benchmarks for enterprise AI agents can pass existing utility checks and still remain structurally unrealistic, especially in privacy-sensitive healthcare settings where operational data are hard to access. We study how to improve such benchmarks without breaking the downstream utility checks already used in practice. We formulate benchmark revision as utility-constrained realism improvement: dataset changes should increase realism while staying above an operational utility floor. We instantiate this idea on a care-gap benchmark derived from Synthea-generated patients exercised through demonstration electronic health record workflows and then processed by the same downstream pipeline as operational data. Realism is measured through missingness structure, simplicity, structural plausibility, and population alignment. The baseline benchmark is extremely thin: sampled-pair missingness is 79.44%, only 12.75% of rows are actionable, 38.94% of patients have zero actionable measures, and top-three token concentration reaches 100.0%. Two deterministic revisions improve these panels while remaining above the current utility floor, whereas a naive densification control preserves unrealistic templating. We further show that internal benchmark realism and source fidelity to an aggregate operational reference are related but distinct objectives. These results suggest that synthetic benchmark quality should be optimized explicitly, with utility treated as one constraint rather than as sufficient evidence of realism.

Omid Bazgir, Md Nasir, Jacob Hoffman +6

Oracle Health and Life Sciences United States

Synthetic Benchmark Generation Benchmark Validity Synthetic Data Evaluation

Aug 5, 2026cs.SE

SciCode-Verified: How Benchmark Defects Underestimated the Scientific-Coding Ability of Language Models

SciCode is the standard measure of the scientific-coding ability of language models: research-level problems that demand both frontier scientific theory and its implementation as working numerical code. It is a component of the Artificial Analysis Intelligence Index and a standing evaluation in government and national-laboratory suites. Yet its scores have recently plateaued: the strongest 2026 models cluster tightly around 60% subproblem accuracy, and a successor model ties its predecessor. We trace this stagnation to defects in the benchmark itself. A per-problem, domain-expert audit of all 65 test problems uncovers 263 defects; 192 of them, spread across 91% of the main problems, cause correct, instruction-following solutions to be wrongly rejected---through non-reproducible gold answers, over-tight tolerances, or self-contradictory specifications. Critically, 78% of these score-suppressing defects require specialized physics or mathematics knowledge to detect, not mere clerical proofreading. We corrected every confirmable defect to produce SciCode-Verified. The corrections add only the specifications a well-posed problem requires, repair grading, and tighten the tests that were too lenient; every change is recorded with its justification and independently re-checked by a second domain expert. We re-evaluate twelve frontier model snapshots on the corrected benchmark and find a substantial recovery: subproblem accuracy rises from 45--60% to 84--98%, and main-problem accuracy from 9--27% to 69--92%. State-of-the-art models are far more proficient in scientific coding than SciCode has suggested---the bottleneck was not model capability, but the quality of the evaluation instrument. We release SciCode-Verified with its complete audit trail as the corrected public standard.

Sihan Hu, Lyuhan Huang, Youjin Deng +1

Hefei National Laboratory College of Mechanical and Electrical Engineering University of Science and Technology of China · Hefei National Laboratory Institute of Theoretical Physics University of Science and Technology of China Chinese Academy of Sciences Hefei 230026, China · Chinese Academy of Sciences Beijing 100190, China

LLM Evaluation Scientific Code Generation Benchmark Validity Code Generation Benchmarks

Aug 5, 2026cs.CL

NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap

We introduce NOLLI, a procedurally generated English-Korean puzzle benchmark designed to diagnose where Korean performance gaps arise. It comprises 15 puzzle types (25 tasks; 7,500 items), with every instance seed-regenerable, verified to have a unique solution, and scored deterministically. Rather than equating harder with bigger, we calibrate difficulty behaviorally, tuning each generator until a fixed reference model lands in target accuracy bands. Its three-level design combines matched direct translations, script adaptations over Hangul jamo (sub-syllabic letters), and Korean-only tasks grounded in Korean culture or orthography. We evaluate 15 frontier, open-weight, and Korean-developed models; among the 12 above a 3% overall-accuracy floor, matched English-Korean accuracy is statistically equivalent within a +/- 10 pp margin (TOST), suggesting little cost from presentation language alone. Writing-system-intensive tasks show sharper gaps: Korean Cipher falls behind English by up to 68.7 pp, whereas Cryptarithmetic over the same jamo shows no systematic penalty, and Jamo Composition accuracy predicts Korean Cipher accuracy. These contrasts are diagnostic rather than causal, consistent with difficulty in multi-step sub-syllabic execution. Korean-only tasks separate rule-application deficits, which vary in sign, from a Kinship deficit positive in all 12. Finally, a salient size measure fails to grow from Easy to Hard in 7 of 15 types, making structural size an unreliable proxy for empirical difficulty.

Dasol Choi, Joonyong Park, Daegon Yu +3

AIM Intelligence · HAERAE LAB · KT Corp. +3

Multilingual Language Model Evaluation Benchmark Validity LLM Reasoning Cross-Lingual Transfer

Aug 4, 2026cs.CL

Benchmarking the Benchmarks: Testing the Predictive Validity of Commonsense Benchmarks

Predicting LLM's capabilities on real-world tasks is essential, yet the extent to which performance on commonsense benchmarks predicts downstream performance remains underspecified. To establish the practical usability of widely adopted commonsense benchmarks, we evaluate 23 models from six families on four established commonsense benchmarks, four reworked variants, three non-commonsense controls, and eight downstream tasks requiring implicit social, pragmatic, temporal, or physical reasoning. We compare model rankings, compute controlled correlations, and use leave-one-family-out cross-validation to assess the criterion validity of commonsense benchmarks. Our results show that revised benchmarks largely preserve original model rankings and do not improve downstream predictive power. Commonsense benchmarks show consistent cross-family predictive validity for only a narrow subset of downstream tasks, with smaller or metric-specific gains elsewhere. Overall, standardized commonsense benchmarks provide task-dependent rather than broad evidence of downstream commonsense competence.

Ine Gevers, Walter Daelemans

CLiPS, University of Antwerp

LLM Evaluation Reasoning Evaluation Implicit Reasoning in Language Models Benchmark Validity

Aug 3, 2026cs.HC

Reassessing the Feasibility of PPG-Based Non-Invasive Blood Glucose Level Estimation

Non-invasive blood glucose level (BGL) estimation from photoplethysmography (PPG) holds great promise for wearable health monitoring, but results across studies are hard to compare due to inconsistent datasets, data leakage, and non-standardized evaluation metrics. We present the first reproducible, extensible evaluation pipeline and use it to reassess five representative PPG-based BGL methods on published datasets under three increasingly strict data-split protocols: random window-level, participant-aware, and leave-some-participants-out (LSPO). Models appeared competitive under random splitting but collapsed under participant-aware and LSPO evaluation, with nearly all yielding near-zero or negative R

^2

values comparable to a mean-prediction baseline. Critically, across every model and split, over 90% of predictions fell within clinically acceptable zones (Clarke Error Grid A+B), including the baseline. This reveals a fundamental disconnect: clinical zone metrics systematically conceal model failure in this domain. Our findings demonstrate that random train-test splits substantially overestimate the generalization of PPG-based BGL models due to sample-level data leakage, and that robust ML evaluation must precede clinical validation to meaningfully assess real-world utility.

Supraja Ramesh, Markus Neufeld, Michael Küttner +2

Karlsruhe Institute of Technology, Karlsruhe, Germany

Data Leakage OOD Generalization Benchmark Validity Photoplethysmography Wearable Health Monitoring

Aug 2, 2026cs.LG

UpliftBench: Revealing Outcome-Regime and Objective Mismatch in Uplift Evaluation

Uplift modeling (conditional-average-treatment-effect estimation) drives personalized targeting, yet published uplift benchmarks frequently disagree on which estimator performs best; we show the disagreement is substantially about metrics, not models. UpliftBench evaluates 12 uplift estimators under an outer-test-isolated, multi-objective protocol across seven dataset families; its two findings are identified where a reference objective exists -- F1 on the standard continuous benchmark (IHDP), F2 in a within-sample case study on Jobs. On that benchmark, Qini shows no detectable alignment with effect accuracy -- across all 100 IHDP realizations its mean rank correlation with effect accuracy is +0.07, 95% CI [-0.03, +0.16] -- while AUUC is consistently more aligned (paired prefix-mean-AUUC-over-Qini gap +0.49 [+0.40, +0.59]; the shipped cumulative-gain AUUC aligns better still, +0.73). On Jobs, ranking metrics are structurally insufficient for a sign-threshold policy because they discard the score level; empirically, within the released split-rotation analysis direct policy-risk selection yields lower benchmark regret than random model selection while Qini, AUUC, and uplift-at-

k

do not (14-15% regret). Calibrating the decision threshold removes 81% of the Qini-selection regret. Both findings are bounded, not universal: F1 is not detected on either validation family (the ACIC and Revenue-Synthetic gaps are both indistinguishable from zero), and F2 vanishes under a budgeted-value objective where rank suffices. UpliftBench releases versioned loaders, fixed protocols, result artifacts, and a reproducible living leaderboard; the public repository accompanies the paper.

Binshuang Li

Independent Researcher USA

Conditional Average Treatment Effect Estimation Benchmark Design Policy Evaluation Benchmark Validity Individualized Treatment Effect Estimation

Jul 30, 2026cs.AI

How Benchmarks Mis-Score Computer-Use Agents

Computer-use agents (CUA) are being deployed to browse the web and operate desktop software, yet their benchmark scores are still commonly produced by brittle scripted oracles. A score is the output of a pipeline in which tasks can be stale, trajectories can omit decisive visual evidence, evaluators can reject valid alternatives, and aggregate reports can hide the cause of failure. We organize these problems into a reliability framework spanning task construction, trajectory observation, scoring, and reporting. We then audit 150 public failure-scored trajectories from five web, enterprise-workflow, and desktop-control benchmarks, find that 15.3% of FAIL verdicts are wrong: 10.7% are evaluator false negatives and 4.7% are broken tasks. For genuine failures, a three-tier diagnostic taxonomy shows that verification/feedback and planning failures dominate execution/grounding errors, while a single scalar success rate can not explain. We connect these findings to newer long-horizon CUA benchmarks and derive stage-specific design rules for CUA evaluation.

Zihan Dong, Zhiyuan Ma, Zekun Wang +5

Georgia Institute of Technology · North Carolina State University · Lenovo AI Technology Center +3

Agent Failure Analysis Computer-Use Agent Benchmarks Benchmark Auditing Computer-Use Agents Benchmark Validity Automated Evaluation