Hao Mao and coauthors present LegoSR for equation discovery

Hao Mao and coauthors present LegoSR for equation discovery

Hao Mao and coauthors present LegoSR, a framework that brings seven equation-discovery solvers into a shared representation and evaluation process. They aim to make solver results easier to compare and let researchers constrain searches with partially known equations. LegoSR Agent chooses solvers and adjusts searches using solver knowledge, intermediate results and lessons from earlier tasks. On noisy ordinary and partial differential equation benchmarks, the authors report that LegoSR Agent recovered nearly as many equations as the union of seven expert-tuned solvers. They also report that it outperformed a generic agent using the same model, tools and budget by up to 29 percentage points. The abstract describes examples in which the agent’s role grows from running a solver to choosing one and building a partially known equation from failed searches. These reported results concern the specified benchmarks; the abstract does not establish performance on other kinds of scientific discovery. It does not explain how equation recovery was measured, which leaves the basis for assessing the benchmark comparisons unclear.

Publication: a7333a1e-ed60-47ed-8eb8-7490560d2934

attribution — pass: The draft attributes LegoSR and its reported results to Hao Mao and coauthors, which matches the arXiv author list. Benchmark results are framed as what the authors report, not as established fact.

Authors:Hao Mao, Xu T. Liu, Peng Zhao, Shuai Lu, Yuntian Chen

summary — uncertain: The draft faithfully restates the abstract, but a non-specialist may not grasp the core idea. It never says plainly that equation discovery means software inferring mathematical equations or laws from data. Central terms are left unexplained: solvers, symbolic regression, ordinary and partial differential equations, and the union of seven expert-tuned solvers as the comparison point. This leaves the development and its significance hard to follow. Smallest repair: add one plain-language clause explaining equation discovery, and add term-library tooltips for symbolic regression and differential equations. The rest of the summary, the title and the question are faithful.

We present LegoSR, a unified framework for algebraic symbolic regression and differential equation discovery.

date — pass: The originalPublished date of 2026-10-24 matches the v1 submission date. It does not use the v2 revision date of 30 Oct or the feed date of 2 Nov.

Submitted on 24 Oct 2026 (v1), last revised 30 Oct 2026 (this version, v2)

support — pass: Each claim is quoted or closely paraphrased from the abstract and keeps its scope: noisy ODE/PDE benchmarks, nearly as many equations as the union of the solvers, and up to 29 points against a generic agent with the same model, tools and budget. The limitations correctly note that the abstract does not define how recovery was measured or show results on other tasks. The draft invents no benefits, costs or forecasts.

outperforms a generic agent with the same model, tools and budget by up to 29 percentage points

media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to reach the source.