Supplementary materials for "Internal Parallelism Breaks Wall-Clock Budget Invariance in Restart-Fair Metaheuristic Benchmarking"
Year: 2026
Type: supplementary-materials
Abstract: Code, data, thermal logs and the full pre-registration for the ICMLSC 2027 paper Internal Parallelism Breaks Wall-Clock Budget Invariance in Restart-Fair Metaheuristic Benchmarking. The study provides, to the authors' knowledge, the first empirical implementation of the restart-fair protocol for time-fair metaheuristic benchmarking: 15,600 measured conditions across five seed-blocked sessions on one fully characterised platform, holding search behaviour constant while varying per-iteration cost three-, six- and twelvefold over seven thread counts, three budget couplings and fifteen seeds. analysis/reproduce_all.py regenerates every table and headline figure in the paper and asserts the condition counts (15,600 → 1,800 → 1,488 → 62 independent problem×seed blocks). Two things to know before loading the data, both documented in the README: data/voided/ holds a voided unpinned session that must not be analysed, and tau.csv has a non-obvious lookup rule. Licensing is split: MIT for code/ and analysis/, CC BY 4.0 for data/ and thermal/. See LICENSE.
Institutions: Far Eastern University, FEU Institute of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225399679