Where does the time go in hypervolume-based environmental selection? A partly matched study of generational and steady-state truncation in pymoo

Replication package for the manuscript . The study compares the steady-state SMS-EMOA in pymoo 0.6.2 with HV-NSGA-II, an NSGA-II variant whose last-front truncation repeatedly deletes the point with the smallest exclusive hypervolume contribution. Problems: DTLZ2, DTLZ4, DTLZ7, WFG4 and two HVAC control problems (three objectives, 30 seeds, about 10,000 evaluations). Measured: wall-clock run time and hypervolume. Contents: code/ (algorithms, drivers, analysis scripts, tagged corrected-hv-nsga2-v1), results/ (per-run logs for the timing benchmark, instrumented profile, lean SMS-EMOA timing, quality runs, factorial and tuning), config/ (frozen HVAC hypervolume reference), requirements.txt, checksums.sha256, README.md, PROVENANCE.md. Environment: Python 3.11.9, pymoo 0.6.2, numpy 2.4.6, scipy 1.17.1, moocore 0.3.2 (read from the lock file; not a reconstruction of the original installation). Provenance notes: timing rows do not record a code revision, the requested 100% processor power state could not be verified, and several logs were added to version control only at deposit time. They are included unchanged with SHA-256 checksums. See PROVENANCE.md. Licences: code MIT; data and results CC-BY-4.0.