# paper-forge daily report — 2026-09-30

169 new papers screened; 10 analyzed in depth; 0 reproduction runs.

## Reproductions

- none today (reimplementation is opt-in: `forge daily --reimplement`)

## Top candidates

### [When Is Coarse Supervision Worth It? Cost-Aware Learning under Unknown Aggregation](https://arxiv.org/abs/2609.36704)
`2609.36704` · score 62.6 · analysis: heuristic · feasibility 8/10

Modern learning systems often acquire supervision at multiple resolutions, trading annotation cost against information content. We study cost-aware two-resolution learning, where expensive fine labels reveal a vector response and cheaper coarse labels reveal a scalar aggregate formed with unknown weights, while the target remains the full response.

Rubric evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("Synthetic"); + small/synthetic data ("Synthetic"); + public benchmarks ("benchmark"); + concrete method ("closed-form"); + formal statement ("bound"); + claims simplicity ("simple"); no public code found

### [Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents](https://arxiv.org/abs/2609.37236)
`2609.37236` · score 62.1 · analysis: heuristic · feasibility 6/10

An agent that uses tools typically responds to what the user explicitly asks, yet completing the task may require information the user never requested. Work on proactive agents mainly studies whether and when an agent should act on its own, not what information it should pursue.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + public benchmarks ("benchmark"); - system/framework (many moving parts) ("agent that uses tools"); + code: https://github.com/dolev31/ProactiveInquirer; + keywords: retrieval; + 13 HF upvotes

### [Copula Active Subspaces I: A Score-Covariance Method for Reduced-Order Non-Gaussian Density Estimation](https://arxiv.org/abs/2609.36142)
`2609.36142` · score 60.0 · analysis: heuristic · feasibility 6/10

In Bayesian inference problems with non-Gaussian observation noise, the posterior is only as accurate as the noise density, and gradient-based samplers need that density and its gradient evaluable pointwise, whether from an explicit expression or from code, and without an inner solve. We propose Copula Active Subspaces (CAS) to represent this noise density.

Rubric evidence: + cheap/closed-form ("closed form"); + concrete method ("closed form"); + formal statement ("bound"); + code: https://github.com/joshuawchen/copula_active_subspaces; + keywords: regularization

### [Retrieval Sensitivity to Identity Signals in Queries](https://arxiv.org/abs/2609.36534)
`2609.36534` · score 58.4 · analysis: heuristic · feasibility 6/10

Dense retrievers decide which documents reach users and the language models that use them, yet they are typically evaluated with neutral queries. We ask whether the identity signals that real users express in their queries---political ideology and dialect---bias what a retriever returns.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + code: https://github.com/Andrewtcr/bias-ret; + keywords: retrieval

### [MoRE: Scaling mixture of experts with hardware-aware low-rank routing](https://arxiv.org/abs/2609.36301)
`2609.36301` · score 57.9 · analysis: heuristic · feasibility 6/10

Mixture-of-Experts (MoE) layers are central to frontier language models, and recent architectures push toward more and smaller experts. In this regime, the standard linear router becomes a bottleneck: with $M$ experts and hidden dimension $h$, its per-token cost $Θ(Mh)$ dominates the MoE layer once $M$ is large.

Rubric evidence: - pretraining ("pretraining"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("We prove"); + claims simplicity ("simple"); + code: https://github.com/Matheart/MoRE_code; + keywords: kernel

### [Beyond Selection: Token Parameterization for Extreme Visual Token Compression](https://arxiv.org/abs/2609.35232)
`2609.35232` · score 57.4 · analysis: heuristic · feasibility 6/10

Visual-token compression is effective for improving the efficiency of vision-language models, but under extreme compression budgets, token pruning can break visual grounding while learned resamplers increase parameter count, attention cost, and training complexity. We revisit compression through a token parameterization lens, separating (i) basis transformation and structured truncation (retained subspace/compressibility) from (ii) coordinate organization (optimization and cross-modal alignment).

Claimed: accuracy 23.0 (abstract (regex extraction, unverified)); accuracy 95.2% (abstract (regex extraction, unverified)); accuracy 84.2% (abstract (regex extraction, unverified)); accuracy 36.0% (abstract (regex extraction, unverified))
Rubric evidence: + CPU-friendly ("lightweight"); + formal statement ("bound"); + claims simplicity ("straightforward"); + code: https://github.com/zrrraa/Braco; + 3 HF upvotes

### [Understanding Private Evolution as Learning-Augmented Clustering](https://arxiv.org/abs/2609.36678)
`2609.36678` · score 57.4 · analysis: heuristic · feasibility 8/10

Private Evolution (PE) is a differentially private algorithm for synthetic data generation. While it can be viewed as a Wasserstein learning algorithm, it performs much better in practice than worst-case Wasserstein analyses would predict.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + explicit algorithm/pseudocode ("Algorithm 2"); + formal statement ("convergence"); + claims simplicity ("simple"); no public code found

### [OTROPE: Optimal Transport-based Robust Off-policy Evaluation for Large Language Models](https://arxiv.org/abs/2609.36264)
`2609.36264` · score 56.3 · analysis: heuristic · feasibility 6/10

Reliable evaluation of large language models (LLMs) is essential for their development and deployment, yet is often costly, risky, and difficult to perform safely online. We study off-policy evaluation for LLMs, where limited human-labeled data from a behavior model are used to evaluate a newer target LLM.

Rubric evidence: - LLM-based ("Large Language Model"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("we propose a new"); + formal statement ("convergence"); + code: https://github.com/LinerXiang/OTROPE

### [Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation](https://arxiv.org/abs/2609.37661)
`2609.37661` · score 54.7 · analysis: heuristic · feasibility 6/10

Graph-based retrieval-augmented generation supports multi-hop retrieval by organizing corpus information into graphs. However, existing relation-free graph retrieval methods rely primarily on query-sentence similarity to search for evidence.

Rubric evidence: + public benchmarks ("benchmarks"); + concrete method ("we propose a simple"); + claims simplicity ("simple"); + code: https://github.com/Jacob-biu/NexusRAG; + keywords: retrieval

### [The Signed Geometry of One-Shot Recourse: On-Path Validity and the Signed-Curvature Criterion](https://arxiv.org/abs/2609.36252)
`2609.36252` · score 53.7 · analysis: heuristic · feasibility 7/10

Closed-form recourse moves a rejected user along the unit gradient $\hat g$ of the classifier score $f$ by the promised distance $d_p=|f(x)|/\|\nabla f(x)\|$, at which the linearized score reaches zero. We ask when this one-shot step succeeds and what additional model queries change.

Rubric evidence: + cheap/closed-form ("Closed-form"); + small/synthetic data ("MNIST"); + concrete method ("Closed-form"); + formal statement ("bound"); no public code found

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 62.6 | [When Is Coarse Supervision Worth It? Cost-Aware Learning under Unknown Aggregation](https://arxiv.org/abs/2609.36704) `2609.36704` | 0.85 | 0.80 | 0.90 | 0.00 | 0.00 | heuristic | True |
| 62.1 | [Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents](https://arxiv.org/abs/2609.37236) `2609.37236` | 0.65 | 0.80 | 0.30 | 1.00 | 0.50 | heuristic | True |
| 60.0 | [Copula Active Subspaces I: A Score-Covariance Method for Reduced-Order Non-Gaussian Density Estimation](https://arxiv.org/abs/2609.36142) `2609.36142` | 0.70 | 0.40 | 0.75 | 1.00 | 0.20 | heuristic | True |
| 58.4 | [Retrieval Sensitivity to Identity Signals in Queries](https://arxiv.org/abs/2609.36534) `2609.36534` | 0.65 | 0.75 | 0.40 | 1.00 | 0.20 | heuristic | True |
| 57.9 | [MoRE: Scaling mixture of experts with hardware-aware low-rank routing](https://arxiv.org/abs/2609.36301) `2609.36301` | 0.40 | 0.80 | 0.70 | 1.00 | 0.20 | heuristic | False |
| 57.4 | [Beyond Selection: Token Parameterization for Extreme Visual Token Compression](https://arxiv.org/abs/2609.35232) `2609.35232` | 0.65 | 0.40 | 0.70 | 1.00 | 0.20 | heuristic | True |
| 57.4 | [Understanding Private Evolution as Learning-Augmented Clustering](https://arxiv.org/abs/2609.36678) `2609.36678` | 0.65 | 0.75 | 1.00 | 0.00 | 0.00 | heuristic | True |
| 56.3 | [OTROPE: Optimal Transport-based Robust Off-policy Evaluation for Large Language Models](https://arxiv.org/abs/2609.36264) `2609.36264` | 0.45 | 0.75 | 0.75 | 1.00 | 0.00 | heuristic | False |
| 54.7 | [Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation](https://arxiv.org/abs/2609.37661) `2609.37661` | 0.50 | 0.45 | 0.75 | 1.00 | 0.20 | heuristic | True |
| 53.7 | [The Signed Geometry of One-Shot Recourse: On-Path Validity and the Signed-Curvature Criterion](https://arxiv.org/abs/2609.36252) `2609.36252` | 0.70 | 0.75 | 0.75 | 0.00 | 0.00 | heuristic | True |
| 50.5 | [Prompts Live on an Arc: Gaussian Curricula in Fisher--Rao Coordinates for Rollout-Efficient GRPO](https://arxiv.org/abs/2609.38018) `2609.38018` | 0.70 | 0.45 | 0.60 | 0.00 | 0.40 | — | — |
| 48.9 | [EngiWorld: What Can Frontier Agents Deliver in Professional Engineering Environments?](https://arxiv.org/abs/2609.37686) `2609.37686` | 0.50 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 48.9 | [Iterative Exact Discrete Guidance for Energy-Based Sampling](https://arxiv.org/abs/2609.37043) `2609.37043` | 0.65 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 48.9 | [LOCO-AdaMP: Built-in LOCO Inference for Adaptive Minipatch Ensembles with Enhanced Prediction](https://arxiv.org/abs/2609.36396) `2609.36396` | 0.65 | 0.75 | 0.45 | 0.00 | 0.20 | — | — |
| 47.9 | [Context Language Models](https://arxiv.org/abs/2609.37725) `2609.37725` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [Cropland PAtteRNS: Parallel Dimensional Attention Networks and Attention to Dataset Disparity for Crop Segmentation in Satellite Imagery Time Series Data](https://arxiv.org/abs/2609.38165) `2609.38165` | 0.50 | 0.40 | 0.60 | 1.00 | 0.00 | — | — |
| 47.4 | [One Proposal for Every Margin: Zero-Shot Amortized Sequential Importance Sampling for Binary Matrices](https://arxiv.org/abs/2609.35514) `2609.35514` | 0.65 | 0.75 | 0.30 | 0.00 | 0.30 | — | — |
| 47.2 | [EmoRES-TTS: Residual-Enhanced Vector Steering for Emotional Speech Generation](https://arxiv.org/abs/2609.38157) `2609.38157` | 0.50 | 0.40 | 0.40 | 1.00 | 0.26 | — | — |
| 46.8 | [Beyond the Timeline: Augmenting Long-Video Memory with Grounded Entity Biographies](https://arxiv.org/abs/2609.38155) `2609.38155` | 0.50 | 0.45 | 0.30 | 1.00 | 0.30 | — | — |
| 46.3 | [Kolmogorov-Arnold Classifier Systems as Universal Approximators](https://arxiv.org/abs/2609.37958) `2609.37958` | 0.50 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 46.3 | [A Finslerian Approach for Embedding Directed Data](https://arxiv.org/abs/2609.37649) `2609.37649` | 0.50 | 0.75 | 0.55 | 0.00 | 0.20 | — | — |
| 46.3 | [Why Adaptive Optimizers Underestimate Rare Tokens](https://arxiv.org/abs/2609.37535) `2609.37535` | 0.70 | 0.40 | 0.60 | 0.00 | 0.20 | — | — |
| 45.8 | [Correct Answers, Invalid Traces: What Verifiable Grade-School Math Reveals About Chain-of-Thought Traces](https://arxiv.org/abs/2609.38107) `2609.38107` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [ReCIRC: Rectified Conformal Risk Control](https://arxiv.org/abs/2609.38112) `2609.38112` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Can a Cacheable Decision Model Follow Rules?](https://arxiv.org/abs/2609.37832) `2609.37832` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Does the Unsafe Gradient Survive a Conversation? On the Fragility of Gradient-Based Jailbreak Detection in Multi-Turn Dialogue](https://arxiv.org/abs/2609.36849) `2609.36849` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [When Noise Estimation Hides Basis Misspecification in Repeated Bayesian Inverse Problems](https://arxiv.org/abs/2609.37762) `2609.37762` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Cheap and Powerful Tests for Supervised Subspaces: Per-Component Inference for PLS](https://arxiv.org/abs/2609.36307) `2609.36307` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.2 | [CaptchaArena: A Large-Scale, Fine-Grained Dataset for Training Computer-Use Agents on Interactive CAPTCHAs](https://arxiv.org/abs/2609.31957) `2609.31957` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 44.1 | [Chinese-Jev: Bringing System One Model to Chinese-Language Tasks](https://arxiv.org/abs/2609.36965) `2609.36965` | 0.40 | 0.40 | 0.40 | 1.00 | 0.26 | — | — |
| 43.7 | [Post-Anomaly Detection Inference for Deep SVDD](https://arxiv.org/abs/2609.37935) `2609.37935` | 0.65 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 43.7 | [Learning When to Update: A Near-Optimal Timing Bandit Approach](https://arxiv.org/abs/2609.37932) `2609.37932` | 0.65 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 43.7 | [High-Dimensional Simulation-Based Inference in Latent Spaces](https://arxiv.org/abs/2609.37381) `2609.37381` | 0.65 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 43.6 | [TabFM: A Zero-Shot Foundation Model for Tabular Data](https://arxiv.org/abs/2609.37959) `2609.37959` | 0.45 | 0.80 | 0.40 | 0.00 | 0.26 | — | — |
| 43.2 | [doPlan: A Variable-Horizon Dataset for Multi-Stage Language-Conditioned Planning in Autonomous Driving](https://arxiv.org/abs/2609.38028) `2609.38028` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 43.2 | [Effective Dense Retrieval using Only In-Context Examples](https://arxiv.org/abs/2609.38099) `2609.38099` | 0.30 | 0.40 | 0.55 | 1.00 | 0.20 | — | — |
| 42.6 | [Beyond Conditional Independence: Root Cause Analysis with Deep Causal Models](https://arxiv.org/abs/2609.36771) `2609.36771` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [StructRL: Online Structured Reinforcement Learning for Long-Horizon Vision-Language-Action Tasks](https://arxiv.org/abs/2609.36352) `2609.36352` | 0.50 | 0.40 | 0.30 | 1.00 | 0.10 | — | — |
| 42.6 | [Omni-IO Skills: Harnessing Your Agent Omni-Native](https://arxiv.org/abs/2609.31847) `2609.31847` | 0.30 | 0.40 | 0.30 | 1.00 | 0.50 | — | — |
| 42.1 | [Tail-Influence Sampling for CVaR Policy Evaluation](https://arxiv.org/abs/2609.38096) `2609.38096` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 42.1 | [PMOPD: Task Ordering, Cycling, and Parameter-Update Subspace Protection in Multi-Teacher On-Policy Distillation](https://arxiv.org/abs/2609.34605) `2609.34605` | 0.65 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 41.6 | [A Spectral Theory of Distortion in LLM Graph Reconstruction: Sharp Bounds and Empirical Characterization](https://arxiv.org/abs/2609.38161) `2609.38161` | 0.45 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 41.1 | [Auditable Long-Term Memory: A Deterministic Retrieval Chain Measured at 479/475 of 500 on LongMemEval-S](https://arxiv.org/abs/2609.38021) `2609.38021` | 0.30 | 0.40 | 0.30 | 1.00 | 0.40 | — | — |
| 41.1 | [GeoOutageBench: Benchmarking Ambiguity-aware, Ontology-grounded Geospatiotemporal KGQA for Multimodal Power Outage and Resilience Analysis](https://arxiv.org/abs/2609.36082) `2609.36082` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 41.1 | [Federated Clustering with Unknown Local and Global Cluster Cardinalities](https://arxiv.org/abs/2609.36762) `2609.36762` | 0.70 | 0.40 | 0.50 | 0.00 | 0.00 | — | — |
| 40.5 | [Physical Muon: Orthogonalization as an Equilibrium Computation](https://arxiv.org/abs/2609.37525) `2609.37525` | 0.65 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 40.0 | [Interacting particle guidance for sampling reward-tilted generative priors](https://arxiv.org/abs/2609.37227) `2609.37227` | 0.30 | 0.40 | 0.60 | 0.00 | 0.60 | — | — |
| 40.0 | [Selecting The Most Informative Tokens in Natural Language Autoencoders](https://arxiv.org/abs/2609.37040) `2609.37040` | 0.25 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 39.5 | [Optimizing VLP-aligned Multimodal Intent Representation with Correct Visual Instantiation for Zero-Shot Composed Image Retrieval](https://arxiv.org/abs/2609.36946) `2609.36946` | 0.65 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 39.5 | [A Sharp Transition in Data Reconstruction under Differential Privacy](https://arxiv.org/abs/2609.37344) `2609.37344` | 0.65 | 0.35 | 0.55 | 0.00 | 0.00 | — | — |
| 38.9 | [Retrieval Capacity of Self-Attention Under Competition](https://arxiv.org/abs/2609.37879) `2609.37879` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [HELIX: Purified and Unified - Rethinking Feature Interaction and Sequence Modeling for Large-Scale Recommendation](https://arxiv.org/abs/2609.37183) `2609.37183` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [ARCagent: An Adaptive Retrieval Calibration Agent for Clinical Question Answering](https://arxiv.org/abs/2609.36392) `2609.36392` | 0.30 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 38.9 | [Counterfactual Probing for Parallel Unmasking with Hidden Forest Structure](https://arxiv.org/abs/2609.37841) `2609.37841` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Pointwise or Pairwise: When Do Pairwise Losses Help Reward Learning, Provably?](https://arxiv.org/abs/2609.37209) `2609.37209` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Optimal Multi-Reward Reinforcement Learning](https://arxiv.org/abs/2609.36486) `2609.36486` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Finite-Sample Theory for Fitted Q-Iteration When Actions Are Functions](https://arxiv.org/abs/2609.36390) `2609.36390` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Stochastic Optimization Under Power-Law Spectra: Tight Bounds and Shuffling Analysis](https://arxiv.org/abs/2609.36271) `2609.36271` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [SPARK: A General Goodness-of-Fit Assessment via Residual Projection](https://arxiv.org/abs/2609.37705) `2609.37705` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 38.4 | [Interpretable intrinsic dimension estimation through componentwise calibration of distance and angle](https://arxiv.org/abs/2609.37114) `2609.37114` | 0.45 | 0.40 | 0.60 | 0.00 | 0.20 | — | — |
| 38.4 | [Graph-Split Bayesian Causal Forest for Spatial Heterogeneous Treatment Effect Estimation](https://arxiv.org/abs/2609.36046) `2609.36046` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 38.4 | [Beyond Dyadic Memory: Interaction-Aware Multimodal Memory with Adaptive Agentic Retrieval for Multi-Party Spoken Conversations](https://arxiv.org/abs/2609.32522) `2609.32522` | 0.50 | 0.40 | 0.30 | 0.00 | 0.50 | — | — |
| 38.3 | [PrismQuant: Optimal Null-Space Rotations for Grouped Quantizers](https://arxiv.org/abs/2609.32429) `2609.32429` | 0.20 | 0.40 | 0.50 | 1.00 | 0.16 | — | — |
| 37.9 | [ReMem: Rethinking Perception and Memory in Long-Context Recommendation Agents](https://arxiv.org/abs/2609.37311) `2609.37311` | 0.50 | 0.40 | 0.50 | 0.00 | 0.20 | — | — |
| 37.9 | [Billiger.de Products: A Bilingual Entity Matching Benchmark](https://arxiv.org/abs/2609.37713) `2609.37713` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.4 | [Mira: Memory-Efficient MoE Inference Using Adaptive Caching and Predictive Expert Staging](https://arxiv.org/abs/2609.38090) `2609.38090` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification](https://arxiv.org/abs/2609.37861) `2609.37861` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Same Bytes, Different Authority: Reserved-Token Representations in Chat-Template Prompt Injection](https://arxiv.org/abs/2609.35932) `2609.35932` | 0.30 | 0.45 | 0.30 | 1.00 | 0.10 | — | — |
| 37.2 | [PlaylistEval: Can Video-Language Judges Be Trusted at Day Scale and Beyond?](https://arxiv.org/abs/2609.34314) `2609.34314` | 0.50 | 0.45 | 0.30 | 0.00 | 0.36 | — | — |
| 36.8 | [Dr. OPD: Learning What to Follow for Optimal On-Policy Distillation of Large Language Models](https://arxiv.org/abs/2609.38025) `2609.38025` | 0.50 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 36.8 | [LongHarness Bench: Stress-Testing Language Model Harnesses for Long-Context Reasoning](https://arxiv.org/abs/2609.38137) `2609.38137` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [BITEM at the NTCIR-19 R2C2 Task: Predicting Confidence from Agentic RAG Pipeline Signals](https://arxiv.org/abs/2609.37993) `2609.37993` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [Towards Semi-Automatically Comparing Keyword-Based and Semantic Search Accuracy](https://arxiv.org/abs/2609.37749) `2609.37749` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [GRP v0.1 Technical Report](https://arxiv.org/abs/2609.36688) `2609.36688` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [Second-Moment Stochastic Approximation Methods](https://arxiv.org/abs/2609.36600) `2609.36600` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.8 | [GNA: Granular Neighbor Assembly for Retrieval-Augmented Multivariate Time-Series Forecasting](https://arxiv.org/abs/2609.36281) `2609.36281` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.3 | [Learning Beyond What You Sample: Off-Policy-Aware Cross-Model Trajectory Exchange for RLVR](https://arxiv.org/abs/2609.37868) `2609.37868` | 0.50 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 35.8 | [Breakdown of Local Denoising as Semantic Speciation](https://arxiv.org/abs/2609.38176) `2609.38176` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Achieving an $O(1/N)$ Optimality Gap in Average-Reward Weakly-Coupled MDPs](https://arxiv.org/abs/2609.38132) `2609.38132` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Dimensionally consistent surrogate modelling through dimensional analysis and harmonic expansions](https://arxiv.org/abs/2609.38094) `2609.38094` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Optimal Quantum-Classical Separations for Exact Learning](https://arxiv.org/abs/2609.38073) `2609.38073` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [The finite-horizon five-expert prediction problem](https://arxiv.org/abs/2609.38035) `2609.38035` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Generated Query Expansion Still Helps Strong Sparse Retrieval: A Controlled Study with SPLADE-v3](https://arxiv.org/abs/2609.37911) `2609.37911` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [A shape-similarity latent space for fluid interfaces: invertible reduced-order modelling of droplet morphology](https://arxiv.org/abs/2609.37947) `2609.37947` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [The double descent and Runge phenomena in overparametrized polynomial interpolation](https://arxiv.org/abs/2609.37657) `2609.37657` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Can Representation Learning Decouple from Loss Minimization? Polar Updates Have an Answer](https://arxiv.org/abs/2609.36240) `2609.36240` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [One-Step Next-Latent Prediction Is Not a World Model](https://arxiv.org/abs/2609.36227) `2609.36227` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.3 | [Improving Function Space Flow Matching with Kernel Optimal Transport](https://arxiv.org/abs/2609.38049) `2609.38049` | 0.35 | 0.45 | 0.55 | 0.00 | 0.20 | — | — |
| 35.3 | [Adapting Linear-Time Architectures for Tabular In-Context Learning](https://arxiv.org/abs/2609.36337) `2609.36337` | 0.25 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 34.7 | [No Scale Left Behind: Multi-Scale Autoencoder with Bi-directional Attention for Time Series Anomaly Detection](https://arxiv.org/abs/2609.38004) `2609.38004` | 0.50 | 0.45 | 0.45 | 0.00 | 0.00 | — | — |
| 34.7 | [TabFM-Auto: Self-Evolving Pipelines for Tabular Foundation Models](https://arxiv.org/abs/2609.37989) `2609.37989` | 0.20 | 0.80 | 0.40 | 0.00 | 0.20 | — | — |
| 34.7 | [Scaffolding Minds: Optimizing Latent Visual Target Representations for Multimodal Reasoning](https://arxiv.org/abs/2608.19669) `2608.19669` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Multi-Agent Flow Matching with Decoupled Generative Guidance](https://arxiv.org/abs/2609.38133) `2609.38133` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Identifiability Guarantees for Drivers and Dynamics of Delayed Physical Systems](https://arxiv.org/abs/2609.37944) `2609.37944` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Thinking in Depth, Speaking Directly: Recurrent Latent Reasoning for Paralinguistically Grounded Spoken Dialogue](https://arxiv.org/abs/2609.37818) `2609.37818` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Which papyrus HTR is good enough? Character-error-rate tolerance of four papyrological tasks on Greek texts](https://arxiv.org/abs/2609.37755) `2609.37755` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Optimal detection of general moment changes: Simultaneous mean and covariance change detection and beyond](https://arxiv.org/abs/2609.36594) `2609.36594` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.2 | [Scalable Diffusion SBI for Compositional Inference under Simulator Misspecification](https://arxiv.org/abs/2609.36950) `2609.36950` | 0.25 | 0.45 | 0.60 | 0.00 | 0.20 | — | — |
| 32.6 | [Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI](https://arxiv.org/abs/2609.38143) `2609.38143` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms](https://arxiv.org/abs/2609.38121) `2609.38121` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [How Local Mixing Encodes Relative Position in Global NoPE Attention](https://arxiv.org/abs/2609.38109) `2609.38109` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Neural topology optimization of ship structures under propulsion machinery vibrations](https://arxiv.org/abs/2609.38089) `2609.38089` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [When do data mixtures improve scaling laws? Insights from high-dimensional regression](https://arxiv.org/abs/2609.38011) `2609.38011` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Mutual Information Constrained Chernoff Bottleneck](https://arxiv.org/abs/2609.37994) `2609.37994` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Learning What to Remember: Long-horizon Counterfactual Memory Optimization](https://arxiv.org/abs/2609.37930) `2609.37930` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [It's Not What the Image Shows: Irrelevant Context Destabilises VLM Judges Without Informing Them](https://arxiv.org/abs/2609.37863) `2609.37863` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Can Vision-Language Models Stay Helpful When Facing Implicit Risks? Intent-Privilege OPSD for Efficient Safety-Helpfulness Alignment](https://arxiv.org/abs/2609.37837) `2609.37837` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Predictive Geometry of Hidden Trajectories in Transformers](https://arxiv.org/abs/2609.37717) `2609.37717` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [When Models Don't Manipulate Manifolds: The Geometry of a Comparison Task](https://arxiv.org/abs/2609.37680) `2609.37680` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Safer Content or Firmer Refusals? A Hybrid Perturbation Defense for Alignment under Harmful Fine-tuning](https://arxiv.org/abs/2609.36862) `2609.36862` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Search Dimension in Unlabeled Projection Pursuit: A Scaling Law for Subspace Restriction](https://arxiv.org/abs/2609.37917) `2609.37917` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [A stochastic subgradient method with optimal failure exponent](https://arxiv.org/abs/2609.37425) `2609.37425` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Into the danger zone: stable extrapolation in high-dimensional function and operator learning](https://arxiv.org/abs/2609.36709) `2609.36709` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Hierarchical Utility Calibration for Structured Multiclass Decisions](https://arxiv.org/abs/2609.36532) `2609.36532` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [AnthroDial: Benchmarking LLM Anthropomorphism in Autonomous Social Interaction](https://arxiv.org/abs/2609.37853) `2609.37853` | 0.45 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 31.6 | [Jaxolotl: A Unified High-Performance Benchmark Suite for LTL-Based Multi-Task RL](https://arxiv.org/abs/2609.38065) `2609.38065` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Pattern Formation in Transformers](https://arxiv.org/abs/2609.37921) `2609.37921` | 0.50 | 0.75 | 0.00 | 0.00 | 0.00 | — | — |
| 31.6 | [MERGE: Multi-LLM Ensemble for Retrieval via Generative Enrichment](https://arxiv.org/abs/2609.37574) `2609.37574` | 0.30 | 0.45 | 0.30 | 0.00 | 0.40 | — | — |
| 31.6 | [Better Nearest Neighbor Graph Indices via (Efficient) LLM-Guided Pruning](https://arxiv.org/abs/2609.36359) `2609.36359` | 0.30 | 0.45 | 0.30 | 0.00 | 0.40 | — | — |
| 31.1 | [The Geometry of Inference in Transformer Residual Streams](https://arxiv.org/abs/2609.37824) `2609.37824` | 0.25 | 0.40 | 0.70 | 0.00 | 0.00 | — | — |
| 31.1 | [Ornstein-Uhlenbeck Is Hard to Beat, Yet Superlinear Drift Ships Lower Transport Costs](https://arxiv.org/abs/2609.37579) `2609.37579` | 0.35 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 31.1 | [EasyPPO: Stabilizing the Critic Is Key](https://arxiv.org/abs/2609.36802) `2609.36802` | 0.30 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 30.5 | [An Efficient Machine Learning Approach for Degradation Forecasting in AEM Water Electrolysis](https://arxiv.org/abs/2609.37941) `2609.37941` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [How Many Labels Does a Language Need? Annotation Budgets and Cross-Lingual Pooling for African-Language Text Classification](https://arxiv.org/abs/2609.37882) `2609.37882` | 0.40 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 30.5 | [ThuRunel: Dynamic Decoupling for Structured Advisory Dialogue](https://arxiv.org/abs/2609.36340) `2609.36340` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [RLTL;DR: Self-improvement by Internalizing Self-generated Feedback](https://arxiv.org/abs/2609.37633) `2609.37633` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Storage Is Not Strategy: State-Conditioned Support Control for LLM Unlearning](https://arxiv.org/abs/2609.37858) `2609.37858` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.4 | [Pretraining Transformers with Quantized Softmax in Attention](https://arxiv.org/abs/2609.33591) `2609.33591` | 0.25 | 0.40 | 0.40 | 0.00 | 0.36 | — | — |
| 30.0 | [Follow the Entities: A Corpus Map for Agentic Search](https://arxiv.org/abs/2609.37226) `2609.37226` | 0.30 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 29.5 | [STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization](https://arxiv.org/abs/2609.38169) `2609.38169` | 0.00 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 29.5 | [AdviSD: Learning to Advise Frontier LLMs via Targeted Multi-Turn Self-Distillation](https://arxiv.org/abs/2609.38142) `2609.38142` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Explore Broadly, Reason Sharply: Push Small Models toward the Frontier via Sampling](https://arxiv.org/abs/2609.38104) `2609.38104` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [KUPAS MASTER: Distilling the Tacit Expertise of Master Practitioners into Agent-Ready Experience Corpora](https://arxiv.org/abs/2609.37673) `2609.37673` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Language Models Are "Insecure" Reporters](https://arxiv.org/abs/2609.36139) `2609.36139` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.9 | [Alpha Diffusion Language Models: Factorization Alone Is Not the Problem](https://arxiv.org/abs/2609.38066) `2609.38066` | 0.35 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 28.4 | [Mnemon: Raw Records, Fast Judgments, Slow Thoughts](https://arxiv.org/abs/2609.36059) `2609.36059` | 0.30 | 0.40 | 0.00 | 1.00 | 0.00 | — | — |
| 27.8 | [Improved Distributional Diffusion Models](https://arxiv.org/abs/2609.37147) `2609.37147` | 0.10 | 0.00 | 0.40 | 1.00 | 0.36 | — | — |
| 27.4 | [Do LLM Agents Execute the Plans They Declare? From Planning-Mode Declaration to Pattern-Specific Execution](https://arxiv.org/abs/2609.38108) `2609.38108` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Traversing the solution space of neural networks with Hessian Null Space Continuation](https://arxiv.org/abs/2609.38081) `2609.38081` | 0.50 | 0.00 | 0.40 | 0.00 | 0.20 | — | — |
| 27.4 | [Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering](https://arxiv.org/abs/2609.38177) `2609.38177` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [The Unequal Influence of Bad Advice: Using Training Data Attribution to Modulate Emergent Misalignment](https://arxiv.org/abs/2609.37914) `2609.37914` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Probe-Space Preconditioning for Fast and Stable Zero-Order Training](https://arxiv.org/abs/2609.38095) `2609.38095` | 0.00 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 26.3 | [Dagger: Decoupling-based Model Stealing Attack against Graph Neural Networks](https://arxiv.org/abs/2609.37972) `2609.37972` | 0.50 | 0.05 | 0.30 | 0.00 | 0.20 | — | — |
| 26.3 | [Pruning for Efficiency, Paying in Fairness: Demographic Disparities in Pruned Speech-LLMs](https://arxiv.org/abs/2609.38106) `2609.38106` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Gender bias across LLMs is common and highly heterogenous](https://arxiv.org/abs/2609.38036) `2609.38036` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning](https://arxiv.org/abs/2609.37915) `2609.37915` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data](https://arxiv.org/abs/2609.37807) `2609.37807` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [On Trajectory-Aware Training for Masked Diffusion Language Models](https://arxiv.org/abs/2609.37974) `2609.37974` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [SelfSearch: Reward-Free Search for Self-Improving Agents](https://arxiv.org/abs/2609.37968) `2609.37968` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [A Proposed Rubric for Evaluating Expressed Clinical Reasoning in Large Language Model Responses](https://arxiv.org/abs/2609.37788) `2609.37788` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Selecting What Matters: Semantic Compression-Guided Selective Pooling for Long-Context Embeddings](https://arxiv.org/abs/2609.37782) `2609.37782` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.7 | [Pretraining Latent Information Feedback Transformers with Teacher Supervision](https://arxiv.org/abs/2609.38149) `2609.38149` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Zero-shot Dependency Parsing with Unsupervised Cross-Lingual Bootstrapping](https://arxiv.org/abs/2609.37883) `2609.37883` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Skill-Space Shooting for Autonomous Robot Policy Improvement](https://arxiv.org/abs/2609.38178) `2609.38178` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [Latent Inference-Time Guidance of Time Series Foundation Models](https://arxiv.org/abs/2609.38058) `2609.38058` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [XU-RS: Explaining Credal Width in Random-Set Language Models](https://arxiv.org/abs/2609.37594) `2609.37594` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [Identifying ODEs from Unstructured Data with Causal Representation Learning](https://arxiv.org/abs/2609.37083) `2609.37083` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 20.5 | [ROSS: Relearning from Self-Generated Rollouts through Selective Supervision](https://arxiv.org/abs/2609.35954) `2609.35954` | 0.00 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 20.0 | [LeapQuant: Efficient Linear Attention with Accurate Recurrent State Quantization](https://arxiv.org/abs/2609.38166) `2609.38166` | 0.00 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 19.5 | [Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders](https://arxiv.org/abs/2609.35210) `2609.35210` | 0.30 | 0.40 | 0.00 | 0.00 | 0.10 | — | — |
| 18.4 | [A foundation model for energy and radiation systems built on heterogeneous scientific interfaces](https://arxiv.org/abs/2609.38067) `2609.38067` | 0.05 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Reader Proficiency Shapes Layer-wise Surprisal Profiles](https://arxiv.org/abs/2609.37688) `2609.37688` | 0.30 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 16.8 | [$S^3$: Spectral Null-Space Swap Makes Reasoning Models Efficient](https://arxiv.org/abs/2609.37976) `2609.37976` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Evaluating and Benchmarking the System One Model Jev](https://arxiv.org/abs/2609.37647) `2609.37647` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 15.8 | [Time-Anchored Diffusion Language Models: Latent-Space Caching for Fast Generation](https://arxiv.org/abs/2609.37924) `2609.37924` | 0.00 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 15.8 | [It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs](https://arxiv.org/abs/2609.37891) `2609.37891` | 0.00 | 0.35 | 0.40 | 0.00 | 0.00 | — | — |
| 14.7 | [Scene-Consistent Illumination Transfer for Inserted Advertising Graphics](https://arxiv.org/abs/2609.37951) `2609.37951` | 0.10 | 0.40 | 0.15 | 0.00 | 0.00 | — | — |
| 14.7 | [From Routing Signals to Selective Review: Visual regrounding in MoE VLMs](https://arxiv.org/abs/2609.38111) `2609.38111` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 13.2 | [Hyperspherical Semantic Trajectory Analysis: Mapping Technological Diffusion across Academic Preprints, Patent Signals, and Compute Scaling](https://arxiv.org/abs/2609.35845) `2609.35845` | 0.15 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |

Scores: weighted mean of rubric criteria (0–1 each) × 100; see README for the rubric.