Code and data for: Beyond expression size – abstract-syntax-tree descriptors predict symbolic integrability and computational cost in a computer algebra system

This package contains the code, data and results for a study of what makes symbolic expressions hard for computer algebra systems. A seeded generator produced 4,481 distinct symbolic expressions. Each was canonicalised with the MATLAB Symbolic Math Toolbox and described by 25 abstract-syntax-tree descriptors and 3 size baselines. For each expression the package records closed-form integrability within 10 s, integration time, derivative growth and simplification time in MATLAB R2025a. An independent replication with SymPy 1.14.0 is included, together with 365 external tasks derived from the AI Feynman equations. Contents: MATLAB code (RUN_ALL.m reproduces the study with a fixed seed) and a Python script for the SymPy replication. Datasets with all descriptors and outcomes. SymPy outcomes. The result tables behind every table and figure of the article. Figures 1–8. A README, a data dictionary and a citation file. The AI Feynman equation list (Udrescu & Tegmark, Sci. Adv. 2020) is not redistributed; Data/README.md explains where to download it.