A Graph-Based Cross-Fitting Engine in R R package crossfit version 0.1.3 Provides a general cross-fitting engine for semiparametric estimation (e.g., double/debiased machine learning). Supports user-defined target functionals and directed acyclic graphs of nuisance learners with per-node training fold widths, target-specific evaluation windows, and fold-allocation modes ("overlap", "disjoint", "independence"). Returns either numeric estimates (mode = "estimate") or cross-fitted prediction functions (mode = "predict"), with configurable aggregation over panels and repetitions, reuse-aware caching, and failure isolation, making it well-suited for simulation studies and large benchmarks. Software Depends: R (>= 4.1.0) Imports: stats, utils Suggests: knitr, rmarkdown, testthat (>= 3.0.0) Etienne Peyrot <etienne.peyrot@inserm.fr> Comprehensive R Archive Network (CRAN) Etienne Peyrot [aut, cre] (ORCID: <https://orcid.org/0009-0006-8520-6201>) GPL-3 2026-03-04 application/tgz https://CRAN.R-project.org/package=crossfit doi:10.32614/CRAN.package.crossfit