An iterated local search for the Book Scanning Problem

Abstract Digitizing a library network's collections before a deadline requires coordinating library sign-up and book assignment, and this plan determines how much of the collections' value is scanned. Sequential sign-up, limited scanning capacities, and overlapping collections make the two decisions interdependent, since activating one library delays the others and each book counts only once. The Book Scanning Problem of Google Hash Code 2020 captures this setting, and its contest solvers are tailored to individual instances. We propose an iterated local search that optimizes the library sign-up order and derives a feasible book assignment from each order with fast decoders. It constructs an initial solution, improves library selection and ordering by local search, and applies perturbations and restarts when progress stalls. Its main parameters are configured automatically, once for all instances. We also provide a formal model and a public benchmark of 324 instances: the official contest instances and three generated families, divided into 256 training and 68 evaluation instances. Ablations trace the solution quality to the local search and identify library insertion and replacement as the operator group with a significant effect. With the best of ten runs per instance, the method scores 2.362\% more in total than the per-instance best of six existing methods and is better on 24, equal on 31, and worse on 13 of the 68 instances. On the five official contest instances it remains competitive, 0.058\% below the reference total. Paired statistical tests confirm the advantage, and a simple upper bound proves 21 solutions optimal.