Robopol Refined is a reusable guided search engine for combinatorial optimization. The paper presents the method as a universal framework built around a generic search core and problem-specific adapters. The core handles candidate management, beam-style selection, repeated improvement, perturbation, deterministic parallel search, and diagnostics, while adapters supply scoring, move generation, guidance edges, local search, and domain-specific validity logic. The document focuses on the public architecture and positioning of Robopol Refined across routing, scheduling, packing, graph, assignment, and subset-selection problems. It intentionally avoids disclosing proprietary implementation details such as exact edge ranking formulas, candidate scoring rules, perturbation schedules, parameter recipes, or benchmark-specific tuning.
Robert Polak (Mon,) studied this question.