Suite and data provenance
The formulas are clean-room Rust implementations from the primary definitions listed below. This directory contains no copied competition implementation, shift vector, rotation matrix, COCO binary, or externally licensed data file.
| Suite | Definition source | Repository policy |
|---|---|---|
| Sphere, Rosenbrock, Rastrigin, Ackley, Griewank, Schwefel, Levy, Zakharov | standard mathematical definitions, stated explicitly in classic.rs | formulas only |
| ZDT1–4, ZDT6 | Zitzler, Deb and Thiele, Comparison of Multiobjective Evolutionary Algorithms, Evolutionary Computation 8(2), 2000, DOI 10.1162/106365600568202 | formulas and analytic fronts |
| DTLZ1–7 | Deb, Thiele, Laumanns and Zitzler, Scalable Test Problems for Evolutionary Multiobjective Optimization, 2005, DOI 10.1007/1-84628-137-7_6 | formulas and analytic fronts |
| WFG1–9 | Huband et al., A Review of Multiobjective Test Problems and a Scalable Test Problem Toolkit, IEEE TEC 10(5), 2006, DOI 10.1109/TEVC.2005.861417 | gated; no unverified implementation ships |
| CEC shift/rotation data | competition distributions | loader and synthetic test fixture only; users obtain data under its own terms |
| BBOB/COCO | Hansen et al., COCO/BBOB documentation | gated; no COCO binary or near-compatible implementation ships |
| Lennard-Jones clusters | University of Cambridge, table of Lennard-Jones cluster minima, with the literature links attached to that table | clean-room pair formula and scalar targets only; no coordinates |
Lennard-Jones target and coordinate policy
The five scalar target values and point-group labels are transcribed from the
linked Cambridge table. They are described as source-cited putative minima.
For the publication artifact, the linked point files were downloaded to a
temporary directory and evaluated by this implementation. All five residuals
are below 4.9e-7; exact source URLs and SHA-256 hashes are retained in
reference-audit.json.
No coordinate file, image, database dump, or structure-derived fixture is
redistributed.
The local loader accepts either conventional XYZ (count, comment, then
symbol x y z) or rows of x y z / symbol x y z. A user may run
--reference-file to evaluate one separately obtained structure, or
--reference-directory during a campaign for files named by atom count.
Artifacts keep reference_structure_audited distinct from target provenance;
a missing file is an error and never selects built-in coordinates.
CEC local-data format
The loader accepts UTF-8 text with comments beginning #:
dimension 2
shift 1.5 -2.0
rotation 1.0 0.0
rotation 0.0 1.0
There must be exactly one shift row and exactly dimension rotation rows,
each with dimension finite values. Missing files, malformed rows, and
non-orthogonal matrices are actionable errors; no identity fallback exists.
Current evidence gates
results/publication/wfg/run.json and results/publication/bbob/run.json
record status: "skipped" and reason: "reference-fixtures-unavailable".
Their schema-v2 manifests also record preset, seed, workers, replay command,
and the fact that implementation was not attempted because independent
fixtures are a prerequisite. This is an evidence result, not a promise that
plausible formulas were tested.
The gate can change only in a later reviewed commit that adds primary-source
fixed-point fixtures with clear redistribution terms.