Code The file hoeft-et-al-icaps2023-code.zip contains our modified version of the Scorpion planner (https://github.com/jendrikseipp/scorpion) which in turn is based on the Fast Downward planning system (http://fast-downward.org). The eager and lazy SPhO implementation is found under "src/search/operator_counting/pho_abstraction_constraints.{h,cc}". In particular, our configurations use the CPLEX 20.1 LP solver (https://www.ibm.com/academic/home). Detailed instructions for compiling the planner can be found online (http://www.fast-downward.org) and for adding the LP support in the file "LPBuildInstructions - Fast Downward Homepage.pdf". To run the configurations used in the paper run: ./fast-downward.py PDDL_TASK --search $configuration where $configuration is: Eager SPhO: "astar(operatorcounting([pho_abstraction_constraints([projections(systematic(2,interesting_general))],strategy=always)],cache_lp=False,debug_cache=False))" Lazy SPhO eqdist: "astar(operatorcounting([pho_abstraction_constraints([projections(systematic(2,interesting_general))],strategy=tuple)],cache_lp=True,debug_cache=False))" Lazy SPhO grouped: "astar(operatorcounting([pho_abstraction_constraints([projections(systematic(2,interesting_general))],strategy=max_cluster)],cache_lp=True,debug_cache=False))" Lazy SPhO range: "astar(operatorcounting([pho_abstraction_constraints([projections(systematic(2,interesting_general))],strategy=range_sa)],cache_lp=True,debug_cache=False))" Lazy SPhO percent: "astar(operatorcounting([pho_abstraction_constraints([projections(systematic(2,interesting_general))],strategy=percent_sa)],cache_lp=True,debug_cache=False))" offline SPhO: "astar(pho_offline([projections(systematic(2,interesting_general))],max_optimization_time=0,max_time=200))" Benchmarks The file ipc-benchmarks-optimal-strips-1998-2018.zip contains the STRIPS PDDL benchmarks from sequential optimization tracks of IPC 1998-2018. Experiment Data: The remaining zipfiles contain the raw experiment data (raw-runs), parsed values and basic reports (parsed-report) for the experiments in the paper. The code directories and benchmark files have been removed to avoid duplication and to save space.

This work was partially supported by TAILOR, a project funded by the EU Horizon 2020 research and innovation programme under grant agreement no. 952215, and by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Al- ice Wallenberg Foundation. The computations were enabled by re- sources provided by the National Academic Infrastructure for Su- percomputing in Sweden (NAISS) and the Swedish National Infras- tructure for Computing (SNIC), partially funded by the Swedish Re- search Council through grant agreements no. 2022-06725 and no. 2018-05973.