NewtonSimAIResearch: Data, Software, and Supplementary Materials for Constrained vs. Unconstrained Free-Fall and Projectile-Motion AI Problem-to-Simulation Generation

This repository supports the pilot study "Constrained vs. Unconstrained Free-Fall and Projectile-Motion AI Problem-to-Simulation Generation." It compares constrained NewtonSimAI simulations (GPT-4o-mini extracts structured question details into a fixed simulation framework) with unconstrained Claude Opus 4.6 generations (the model handles both question interpretation and simulation-code generation). Five Force Concept Inventory questions were evaluated, one generation per question per condition, scored against a 17-criterion technical and physical accuracy rubric. Contents: the reported summary scores (scoring_data.csv); a derived criterion-level scoring file (data/criterion_level_scoring.csv); the unconstrained simulation artifacts; the NewtonSimAI source code; prompts; and research documentation. Note: the criterion-level CSV is reconstructed deterministically from the reported summary scores and the published rubric — it is not an original contemporaneous scoring sheet. Licensing: original source code is MIT; original data and documentation are CC BY 4.0 (see LICENSE_SCOPE.md). The single license field on this record is a simplification of that file-level split. Third-party material: Force Concept Inventory questions/images are third-party and are not redistributed here (see THIRD_PARTY_NOTICES.md).