NeurIPS 2026 — measured-subset corpus atlas
The inventory contains 106 explicit accepted-oral records, including six with only a poster event so far. Statistics cover only actually inspected public author manuscripts. They do not represent every accepted paper or the accepted manuscript layouts.
A figure count means a distinct caption label, not a panel, image or graph count. A six-panel figure counts once. There is no figure quota or implied acceptance threshold. Use the type guides to understand each artifact’s purpose and the gallery to see selected styles directly.
68 / 106inspected PDFs / oral inventory records
2,209physical pages in those PDFs
7editorial contribution types
0separate files, excluded from totals
What the measured manuscripts look like
Observed extraction results, not a target length or figure quota. The middle 50% is the interval from the 25th to the 75th percentile. Missing boundaries are excluded from that span’s denominator, not replaced by zero. Caption counts are extraction-assisted and can miss nonstandard or image-only labels; inspect each paper’s notes.
| Measure | Median | Middle 50% | Full range | Denominator |
|---|---|---|---|---|
| Inspected PDF lengthphysical pages | 29 | 22–35 | 10–167 | 68 papers |
| Before references / appendixestimated page-equivalents | 9.79 | 9.18–10.79 | 8.04–38.47 | 68 papers |
| Attached appendix spanestimated page-equivalents | 13.96 | 8.91–18.91 | 1.96–145 | 62 papers |
| Main figurescaption labels | 5 | 3–6 | 0–12 | 68 papers |
| Main tablescaption labels | 2 | 0–5 | 0–13 | 68 papers |
| Appendix figurescaption labels | 3 | 0.75–5 | 0–123 | 68 papers |
| Appendix tablescaption labels | 4 | 0–8.25 | 0–216 | 68 papers |
| Separate source-file lengthwhole-file physical pages, possibly including repeated main text; not appendix-only length | Not resolved | Not resolved–Not resolved | Not resolved–Not resolved | 0 papers |
| Separate source-file figureswhole-file caption labels; may include overlapping main-text displays | Not resolved | Not resolved–Not resolved | Not resolved–Not resolved | 0 papers |
| Separate source-file tableswhole-file caption labels; may include overlapping main-text displays | Not resolved | Not resolved–Not resolved | Not resolved–Not resolved | 0 papers |
Compare like contributions
Each measured paper has one primary editorial type so the totals sum to 68. Secondary contributions were not separately coded; empty secondary lists do not imply the absence of mixed contributions. These are classifications of the annotated argument, not official NeurIPS tracks. The table uses the same extraction method and shows medians; the category guides below explain what these artifacts should accomplish.
| Primary contribution | Papers | Main figures / tables | Body span ≈ pp. | Appendix span ≈ pp. |
|---|---|---|---|---|
| Architecture & representation | 11 | 5 / 4 | 9.53 | 6.1310/11 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Training, objectives & optimization | 10 | 5 / 4.5 | 9.83 | 13.2910/10 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Theory & formal analysis | 19 | 3 / 0 | 10.55 | 17.9117/19 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Empirical analysis, diagnosis & auditing | 7 | 5 / 1 | 9.39 | 9.935/7 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Datasets, benchmarks & measurement | 10 | 6.5 / 1 | 9.64 | 17.459/10 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Efficiency, inference & systems | 5 | 4 / 6 | 9.2 | 23.915/5 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
| Domain-grounded methods & physical systems | 6 | 5 / 3 | 9.55 | 14.416/6 attached boundaries; separate PDF median Not resolved pp. (0 papers) |
Search all 68 measurement records
Open a paper’s ledger for every detected top-level section, its estimated length and share of the PDF, and figure/table locations. Its existing study explains what those visuals and sections do.
Showing 68 of 68 records.
| Paper / category | PDF pages | Main F / T | Appendix F / T | Body / appendix span ≈ pp. | Notes |
|---|---|---|---|---|---|
| AgentAbstain: Do LLM Agents Know When Not to Act?Datasets, benchmarks & measurement | 56 | 7 / 1 | 6 / 15 | 9.23 / 37.91 | No automatic flagsSource correction retained |
| The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-PlayTraining, objectives & optimization | 23 | 3 / 4 | 0 / 5 | 11.09 / 8.91 | No automatic flagsSource correction retained |
| BEAST3D: animal behavioral analysis and neural encoding from multi-view video via Gaussian splattingDomain-grounded methods & physical systems | 31 | 5 / 0 | 4 / 8 | 10.09 / 15.91 | No automatic flagsSource correction retained |
| PhaseLoRA: Control-Regime-Conditioned Low-Rank Adaptation for Continuous-Action Vision-Language-Action PoliciesArchitecture & representation | 19 | 3 / 4 | 2 / 0 | 9.53 / 6.36 | No automatic flagsSource correction retained |
| HABIT: Human-Aware Behavior and Interaction Training Dataset for Robot ManipulationDatasets, benchmarks & measurement | 30 | 9 / 0 | 17 / 3 | 10.09 / 16.91 | No automatic flagsSource correction retained |
| Fractional State Space Transition for Long Sequence ModelingArchitecture & representation | 28 | 4 / 4 | 0 / 6 | 9.21 / 11.91 | No automatic flagsSource correction retained |
| TiRex-2: Generalizing TiRex to Multivariate Data and StreamingArchitecture & representation | 31 | 6 / 1 | 3 / 6 | 11.07 / 14.93 | No automatic flagsSource correction retained |
| Resilience Matters for Embodied Agents System: New Metrics, Systematic Evaluation, and OptimizationEmpirical analysis, diagnosis & auditing | 12 | 5 / 3 | 0 / 0 | 9.09 / Unresolved | No attached appendix boundary detectedSource correction retained |
| GeoSym127K: Scalable Symbolically-verifiable Synthesis for Multimodal Geometric ReasoningDatasets, benchmarks & measurement | 35 | 5 / 6 | 12 / 11 | 9.09 / 20.91 | No automatic flagsSource correction retained |
| Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf ModelsEfficiency, inference & systems | 35 | 3 / 2 | 4 / 8 | 9.2 / 23.91 | No automatic flagsSource correction retained |
| Proactive Instance Navigation with Comparative Judgment for Ambiguous User QueriesDomain-grounded methods & physical systems | 21 | 5 / 7 | 5 / 8 | 9.77 / 8.91 | No automatic flagsSource correction retained |
| SyncWorld: Visual Calibration Enables World Models as Zero-Shot SimulatorsArchitecture & representation | 24 | 5 / 5 | 4 / 4 | 9.18 / 9.91 | No automatic flagsSource correction retained |
| Benchmark Health Index: A Systematic Framework for Benchmarking the Benchmarks of LLMsDatasets, benchmarks & measurement | 42 | 7 / 1 | 1 / 6 | 17.11 / 12.89 | No automatic flagsSource correction retained |
| Exact Gaussian Moment Matching for Residual Networks: a Second-Order MethodTheory & formal analysis | 167 | 3 / 2 | 123 / 216 | 11.09 / 145 | No automatic flagsSource correction retained |
| Self-Programmed Execution for Language-Model AgentsEfficiency, inference & systems | 75 | 4 / 0 | 1 / 16 | 9.09 / 62.91 | No automatic flagsSource correction retained |
| Non-Colliding Biometric Identities for Digital Entities: Geometry, Capacity, and Million-Scale Virtual Identity ProvisioningDomain-grounded methods & physical systems | 25 | 5 / 3 | 6 / 1 | 9.09 / 12.91 | No automatic flags |
| Catch-Only-One: Non-Transferable Examples for Model-Specific AuthorizationDomain-grounded methods & physical systems | 25 | 3 / 5 | 5 / 5 | 11.09 / 10.36 | No automatic flags |
| DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta DebuggingEmpirical analysis, diagnosis & auditing | 10 | 2 / 2 | 0 / 0 | 8.32 / Unresolved | No attached appendix boundary detectedSource correction retained |
| Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine LearningTraining, objectives & optimization | 48 | 0 / 6 | 5 / 15 | 9.21 / 34.91 | No automatic flagsSource correction retained |
| Query Lower Bounds for Diffusion SamplingTheory & formal analysis | 29 | 3 / 0 | 0 / 0 | 12.39 / 11.91 | No automatic flagsSource correction retained |
| Déjà View: Looping Transformers for Multi-View 3D ReconstructionArchitecture & representation | 20 | 4 / 5 | 5 / 3 | 9.47 / 4.91 | No automatic flagsSource correction retained |
| MOVEBENCH: A Benchmark for Global-Scale Wildlife Movement ForecastingDatasets, benchmarks & measurement | 72 | 4 / 2 | 1 / 8 | 10.09 / 34.91 | No automatic flagsSource correction retained |
| Compact SO(3) Equivariant Atomistic Foundation Models via Structural PruningEfficiency, inference & systems | 27 | 3 / 8 | 2 / 16 | 9.09 / 13.91 | No automatic flagsSource correction retained |
| Mechanism-Aware Ensemble Conditioning for Data-Limited Emulation of Extreme EventsDomain-grounded methods & physical systems | 42 | 5 / 3 | 6 / 10 | 9.34 / 23.9 | No automatic flagsSource correction retained |
| Surprises in Proper Positive-Only LearningTheory & formal analysis | 38 | 0 / 0 | 0 / 0 | 13.09 / 20.91 | No automatic flagsSource correction retained |
| C3P: Contrastive promoter-protein pretraining yields representations capturing bacterial gene regulationTraining, objectives & optimization | 23 | 6 / 2 | 3 / 5 | 10.09 / 7.91 | No automatic flagsSource correction retained |
| Uniform-in-Time Weak Propagation of Chaos in Shallow Neural NetworksTheory & formal analysis | 46 | 2 / 0 | 3 / 0 | 27.81 / 14.47 | No automatic flagsSource correction retained |
| Geometry-Aware Similarity Metrics for Neural Representations on Riemannian and Statistical ManifoldsEmpirical analysis, diagnosis & auditing | 13 | 8 / 0 | 0 / 0 | 8.13 / 2.92 | No automatic flagsSource correction retained |
| Bridging Compute- and Data-Optimal PretrainingEmpirical analysis, diagnosis & auditing | 19 | 8 / 1 | 4 / 6 | 11.09 / 4.91 | No automatic flagsSource correction retained |
| Prediction Under Imperfect Compression: A Theory of Approximate MDLTheory & formal analysis | 26 | 0 / 1 | 0 / 0 | 10.65 / 13.91 | No automatic flagsSource correction retained |
| The Stability of Online Algorithms in Performative PredictionTheory & formal analysis | 18 | 2 / 1 | 0 / 0 | 15.09 / Unresolved | No attached appendix boundary detectedSource correction retained |
| Can In-Context Learning Support Intrinsic Curiosity?Theory & formal analysis | 32 | 4 / 0 | 3 / 0 | 9.66 / 17.91 | No automatic flagsSource correction retained |
| Generative Modeling via DriftingTraining, objectives & optimization | 28 | 4 / 7 | 11 / 4 | 8.11 / 16.96 | No automatic flagsSource correction retained |
| Clarification as Supervision: Reinforcement Learning for Vision-Language InterfacesTraining, objectives & optimization | 22 | 3 / 5 | 1 / 1 | 9.16 / 8.89 | No automatic flagsSource correction retained |
| MaxSketch: Robust Distinct Counting in Streams via Random ProjectionsTheory & formal analysis | 30 | 1 / 2 | 1 / 0 | 9.09 / 12 | No automatic flagsSource correction retained |
| LatentUM: Unleashing the Potential of Interleaved Cross-Modal Reasoning via a Latent-Space Unified ModelArchitecture & representation | 26 | 8 / 4 | 5 / 0 | 13.58 / 6.92 | No automatic flagsSource correction retained |
| Filtered Conformal Ellipsoids for Graph-Native Time SeriesTheory & formal analysis | 45 | 2 / 3 | 1 / 22 | 10.33 / 32.91 | No automatic flagsSource correction retained |
| Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation MonitorsEmpirical analysis, diagnosis & auditing | 30 | 8 / 1 | 9 / 3 | 11.26 / 14.88 | No automatic flagsSource correction retained |
| Pruning and Distilling Mixture-of-Experts into Dense Language ModelsEfficiency, inference & systems | 40 | 8 / 7 | 0 / 12 | 13.34 / 23.91 | No automatic flagsSource correction retained |
| Free Decompression with Algebraic Spectral CurvesTheory & formal analysis | 62 | 6 / 1 | 12 / 4 | 10.09 / 48.86 | No automatic flagsSource correction retained |
| Thinking with Images as Continuous Policy: Numerical Visual Chain-of-ThoughtTraining, objectives & optimization | 12 | 5 / 3 | 1 / 0 | 8.19 / 1.96 | No automatic flagsSource correction retained |
| Impossibility of Distribution-Free Predictive Inference for Individual Treatment EffectsTheory & formal analysis | 30 | 5 / 0 | 1 / 16 | 9.09 / 18.91 | No automatic flagsSource correction retained |
| Representation Fréchet Loss for Visual GenerationTraining, objectives & optimization | 30 | 7 / 4 | 8 / 11 | 10.23 / 16.91 | No automatic flagsSource correction retained |
| ExComm: Exploration-Stage Communication for Error-Resilient Agentic Test-Time ScalingArchitecture & representation | 17 | 2 / 6 | 2 / 1 | 9.09 / 5.91 | No automatic flagsSource correction retained |
| ParetoSlider: Diffusion Models Post-Training for Continuous Reward ControlTraining, objectives & optimization | 26 | 9 / 0 | 7 / 9 | 10.11 / 12.95 | No automatic flagsSource correction retained |
| From Average Sensitivity to Small-Loss Regret Bounds under Random-Order ModelTheory & formal analysis | 51 | 0 / 0 | 0 / 0 | 10.69 / 36.91 | No automatic flagsSource correction retained |
| BrainWorld: A Structural-Prior-Conditioned Generative Model for Whole-Brain 4D fMRI DynamicsDomain-grounded methods & physical systems | 33 | 4 / 2 | 8 / 6 | 9.09 / 18.91 | No automatic flagsSource correction retained |
| Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency OperationsDatasets, benchmarks & measurement | 13 | 9 / 6 | 0 / 0 | 9.8 / Unresolved | No attached appendix boundary detectedSource correction retained |
| SALART-VQA: Diagnosing Whether VLMs Understand Salient Artifacts in Generated ImagesDatasets, benchmarks & measurement | 23 | 6 / 1 | 1 / 6 | 9.24 / 10.3 | No automatic flags |
| Tessellations of Semi-Discrete Flow MatchingTheory & formal analysis | 28 | 6 / 0 | 1 / 0 | 10.55 / 15.66 | No automatic flagsSource correction retained |
| Scale-Sensitive Shattering: Learnability and Evaluability at Optimal ScaleTheory & formal analysis | 32 | 1 / 0 | 0 / 0 | 29.09 / Unresolved | No attached appendix boundary detectedSource correction retained |
| MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual DecodingArchitecture & representation | 20 | 5 / 6 | 5 / 9 | 10.09 / 5.91 | No automatic flagsSource correction retained |
| SAM 3D Animal: Promptable Animal 3D Reconstruction from Images in the WildArchitecture & representation | 16 | 7 / 3 | 3 / 1 | 9.09 / 2.91 | No automatic flags |
| DD-Ranking: Rethinking the Evaluation of Dataset DistillationDatasets, benchmarks & measurement | 21 | 3 / 7 | 0 / 13 | 9.24 / 6.91 | Table numbering discontinuity; see source noteSource correction retained |
| Do We Need Asynchronous SGD? On the Near-Optimality of Synchronous MethodsTheory & formal analysis | 29 | 6 / 0 | 11 / 1 | 8.04 / 18.91 | No automatic flagsSource correction retained |
| Learning Reveals Invisible Structure in Low-Rank RNNsTheory & formal analysis | 30 | 4 / 0 | 8 / 0 | 9.26 / 16.91 | No automatic flags |
| Binding Multiple Modalities via Multimodal Wasserstein BarycenterTraining, objectives & optimization | 35 | 5 / 6 | 3 / 18 | 9.57 / 14 | No automatic flagsSource correction retained |
| TransmissiveGS: Residual-Guided Disentangled Gaussian Splatting for Transmissive Scene Reconstruction and RenderingArchitecture & representation | 15 | 10 / 4 | 0 / 0 | 11.09 / Unresolved | No attached appendix boundary detectedSource correction retained |
| Unifying Goal-Conditioned RL and Unsupervised Skill Learning via Control-MaximizationTheory & formal analysis | 35 | 5 / 0 | 3 / 0 | 9.18 / 21.91 | No automatic flagsSource correction retained |
| Computer Use at the Edge of the Statistical PrecipiceEmpirical analysis, diagnosis & auditing | 24 | 3 / 2 | 5 / 5 | 9.39 / 12.93 | No automatic flagsSource correction retained |
| Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning ModelsTraining, objectives & optimization | 60 | 12 / 13 | 0 / 9 | 38.47 / 13.63 | No automatic flagsSource correction retained |
| Offloading Score: Measuring AI Reliance through Counterfactual WorkflowsDatasets, benchmarks & measurement | 37 | 6 / 0 | 15 / 1 | 9.47 / 23.91 | No automatic flags |
| Colored Noise Diffusion SamplingEfficiency, inference & systems | 31 | 6 / 6 | 2 / 4 | 10.09 / 16.91 | No automatic flagsSource correction retained |
| LittleLearner: Language Models Under Pedagogically-Controlled Knowledge ExposureDatasets, benchmarks & measurement | 32 | 9 / 1 | 8 / 6 | 10.09 / 17.45 | No automatic flagsSource correction retained |
| Testable and Actionable Calibration for Full Swap RegretTheory & formal analysis | 38 | 3 / 2 | 0 / 0 | 30.54 / 3.27 | No automatic flagsSource correction retained |
| Quantized Reasoning Models Think They Need to Think Longer, but They Do NotEmpirical analysis, diagnosis & auditing | 22 | 5 / 1 | 5 / 11 | 9.47 / 9.93 | No automatic flagsSource correction retained |
| A multi-scale information geometry reveals the structure of mutual information in neural populationsTheory & formal analysis | 22 | 5 / 0 | 1 / 1 | 10.09 / 8.91 | No automatic flagsSource correction retained |
| Consistent 3D Surface Flow Model with Global StateArchitecture & representation | 22 | 6 / 5 | 1 / 0 | 10.42 / 4.91 | No automatic flagsSource correction retained |
No matching papers. Reset or use a shorter query.
Measurement method
Section titled “Measurement method”Public author manuscripts can use single- or two-column formats, contain long technical reports, and differ from accepted revisions. The extractor uses page geometry to estimate reading order. All 68 inspected manuscripts received a source-linked heading and boundary audit. Corrections distinguish real section starts from contents entries, bibliography initials and diagram labels, and retain the source’s actual column layout. Unusual quantities can remain unresolved.
Each top-level section extends from its detected heading to the next detected heading. Its share is relative to the entire inspected PDF, not the conference content-page allowance. These are approximate layout envelopes, including whitespace and floats—not exact text lengths, word counts, occupied-ink area or official page-limit accounting. The title, abstract and teaser may precede the first numbered section.
Missing appendix boundaries remain unresolved rather than zero. Separate files have separate denominators. Where a checklist boundary is resolved, it ends the preceding appendix span; checklist pages are not silently treated as technical appendix content. Source revisions, extraction notes and URL-linked corrections remain in each paper ledger.
Not every caption was hand-counted. Counts are extraction-assisted; nonstandard or image-only captions, repeated labels, contents pages and atypical text layers can need corrections. The source page remains authoritative. Category assignments are editorial labels from the writing studies, not official conference tracks; secondary contributions were not separately coded.
Selected examples do not establish visual-style prevalence. An accepted-only collection cannot identify which writing choices caused acceptance. Missing-source and author-version selection add further bias; do not generalize these subset distributions to all 106 papers.
Download structural metadata · Inspect source status and studies.