Reproducibility package: Generic metaheuristic support optimization for hardware-aware multiplierless FIR filter design

Frozen data, author-owned code and hardware evidence supporting the article "Generic metaheuristic support optimization for hardware-aware multiplierless FIR filter design". Optimization campaign. 2,160 completed runs and 21.6 million function evaluations over linear-phase FIR filters of order 60, 90 and 120, under a common interface in which a metaheuristic proposes support priorities, a minimax linear program repairs the coefficients, and a Q1.15 / canonical signed-digit stage produces a multiplierless realization. Three studies with disjoint 30-seed blocks: Sensitivity — 810 runs, seeds 71001–71030 Mediator ablation — 270 runs, seeds 72001–72030 External comparison of twelve methods — 1,080 runs, seeds 73001–73030 Every method sharing an order and seed receives byte-identical initial populations. Each run retains its manifest, initial population, complete per-evaluation history, final checkpoint and coefficients. Repaired-support landscape. Layer-wise feasibility, improvement yield and method diagnostics derived from 2,570,131 within-run-distinct support evaluations, including an audit in which none of the 898,200 uniform random proposals generated after the shared initial populations produced a feasible design after quantization. Hardware evidence. Ten post-fit Cyclone IV realizations, five Quartus fitter seeds each (50 fits) with timing and activity-based power reports, bit-exact RTL simulation, and four bitstreams validated physically on an EP4CE10E22C8N through autonomous built-in self-test, checking 742 outputs bit-exactly and their initiation intervals. Provenance. A SHA-256 inventory covers every distributed file (15,630 entries). Every runner and configuration hash recorded by the 2,160 run manifests has an exact byte-for-byte source snapshot, mapped in manifests/EXECUTED_SOURCE_MAP.md. Structure. Uploaded as three archives so that only what is needed has to be retrieved: 01_core.tar.gz — 81 MB download, 223 MB extracted: code, executed-source snapshots, statistical results, derived landscape data, hardware evidence and references. 02_run_histories.tar.gz — 238 MB download, 2.1 GB extracted: adds the complete per-evaluation histories, coefficients, initial populations and manifests of all 2,160 runs. 03_final_checkpoints.tar.gz — 2.0 GB download, 17.9 GB extracted: adds the per-evaluation support and coefficient records of every run. Every quantitative claim in the article is verifiable from the first two archives. Extract them over a common root; the SHA-256 inventory then verifies the result, and scripts/audit_release.py performs a structural audit. SHA-256 of the uploaded archives: dc847cbb35817b7b7da0cb26b3474889bba13d6958a035c266b7d48e788626b4 01_core.tar.gz c9295f606cabfd25281cb5da4dd5ccc06f94c2f45a3417f65c0dfd29c4d26a08 02_run_histories.tar.gz 3f2238e818beccd7633c075320ec726a7a8946d7e850016d8446e5c7d545d089 03_final_checkpoints.tar.gz Scope and exclusions. Third-party optimizer implementations are not redistributed. This applies uniformly to all nine external competitors — DE, PSO, HBA, MPA, GWO, RIME, APO, COA-Crayfish and SFOA — including the official Marine Predators Algorithm implementation. Result directories and run manifests name those algorithms and record the SHA-256 hash of each file as executed, so the interfaces can be identified without shipping third-party sources. The only optimizer distributed here is the author-owned MPHBS. The article itself is not part of this deposit. The manuscript source, its PDFs, and the figures and tables rendered into them are excluded and published through the journal. The derived data behind every figure and table is retained, and the public scripts regenerate them from the frozen evaluation histories. MPHBS is archived and licensed separately (doi:10.5281/zenodo.21326361) and should be cited independently when used. License. Author-owned source code: MIT. Data, results and hardware evidence: CC BY 4.0. LICENSE-DATA states which files fall under each. Neither license grants any right in third-party material