ScoringBench: A Benchmark for Evaluating Tabular Foundation Models with Proper Scoring Rules

TL;DR

ScoringBench is an open benchmark that evaluates tabular regression models using proper scoring rules to better assess their probabilistic predictions across diverse datasets.

Contribution

It introduces a comprehensive, community-driven benchmark with multiple scoring protocols for evaluating probabilistic tabular models, highlighting the importance of metric choice.

Findings

Model rankings vary significantly depending on the scoring rule used.

Models excelling in point-estimate metrics may perform poorly on probabilistic scores.

Evaluation metrics influence model selection and deployment decisions.

Abstract

Tabular foundation models such as TabPFN and TabICL already produce full predictive distributions, yet prevailing regression benchmarks evaluate them almost exclusively via point-estimate metrics (RMSE, $R^2$). This discards precisely the distributional information these models are designed to provide - a critical gap for high-stakes domains where not all kinds of errors are equally costly. We introduce ScoringBench, an open and extensible benchmark that evaluates tabular regression models under a comprehensive suite of proper scoring rules - including CRPS, CRLS, interval score, energy score, and weighted CRPS - alongside standard point metrics. ScoringBench covers 97 regression datasets from diverse domains, supports transparent community contributions via a git-based leaderboard, and provides two complementary ranking protocols: an ordinal Demsar/autorank approach and a magnitude-preserving z-score ranking approach. Evaluating several models - spanning in-context learners, fine-tuned foundation models, gradient-boosted trees, and MLPs - we find that model rankings shift substantially depending on the scoring rule: models that excel on point-estimate metrics can rank poorly on probabilistic ones, and the top-performing model under one proper scoring rule may rank noticeably lower under another. These results demonstrate that the choice of evaluation metric is not a technicality but a modelling decision - and, for applications where e.g. tail errors are disproportionately costly, a domain-specific requirement with direct consequences for model deployment.