ARROWDE: An Adaptive Rotation-Safe Routed Operator Workflow for Differential Evolution This archive contains the reference MATLAB implementation of ARROWDE and the scripts that reproduce every result reported in the accompanying manuscript, submitted to Scientific Reports. ARROWDE is a differential evolution framework that routes each trial vector to one of six single-evaluation search operators. Three are exploratory — a heavy-tailed long jump, an archive-anchored expansion, and an escape operator — and three are exploitative: an archive-guided pbest mutation, a covariance-aligned refinement, and an eigenvector-aligned precision refinement. Selection combines an upper-confidence-bound bandit with landscape-derived priors computed from population diversity and stagnation. Perturbations are drawn in a rotated frame maintained by an implicit Givens chain, which removes the dependence on the problem's coordinate basis without forming or decomposing a covariance matrix. Parameter adaptation follows the success-history scheme of SHADE, population size is reduced across the run, and a governance layer of evaluation-share caps and a starvation floor bounds how much of the budget any single operator can consume. A success-gated bounded local search refines the incumbent solution. Contents The ARROWDE implementation: the six operators, the routing mechanism, the rotation-safe statistics layer, the evaluation-share safeguards, and the local search Driver scripts for the CEC 2014 and CEC 2017 benchmark suites at 30 and 50 dimensions, the constrained engineering design problems, and the systems of nonlinear equations The ablation variants used in the component analysis The parameter file corresponding to Table 4 of the manuscript Random seeds and the raw per-run output from which every table and figure in the manuscript is generated Requirements. MATLAB R2018a or later. No toolboxes beyond the base installation are required.