METAFOR normalized ELA features: MIXTURE, CEC05, CEC14 and SOCO

Version 3.0.1 replaces six CEC05 designs: F24 and F25 at 10D, 30D and 50D. The original Sobol coordinates were re-evaluated with the restored Sphere-component noise. The complete CEC05 native archive and its raw, scaled and aggregate feature tables incorporate the correction. No separate supplement is required. The other 69 CEC05 raw feature rows and all MIXTURE, CEC14 and SOCO data are unchanged. The per-design metadata record the evaluator fingerprints. Version 3.0.0 remains available at DOI 10.5281/zenodo.23913474. This release contains a normalized exploratory landscape analysis computation for 348 METAFOR problem-dimension identities: MIXTURE 150, CEC05 75, CEC14 66 and SOCO 57, at 10D, 30D and 50D. All 12,876 predictor cells in the common 37-feature core are finite, without imputation. Each identity has one scrambled Sobol design, with 2,048, 4,096 or 8,192 points respectively (the largest power of two not exceeding 250D), RRMset 7, and recorded sampling/evaluation and feature-extraction seeds. The four sampling campaigns used 1,662,976 objective evaluations. Native coordinates and raw objective values are included in the companion archives. Features are computed from unit-box X and per-design sample-minmax Y, using R 4.3.3 and flacco 1.8. The common normalized-ic-grid-v1 contains zero plus 1,751 logarithmically spaced positive epsilons from 1e-20 through 1e15. This resolves the three below-default-grid ic.eps.ratio cases without dropping features. CEC05 F7 and F25 remain unbounded benchmark functions. Their ELA sampling windows are explicitly the report-prescribed initialization regions [0,600]^D and [2,5]^D; these exclude the global optima. Other CEC05 domains are unchanged. CEC14 covers the 22 functions implemented in METAFOR, not all competition functions. MIXTURE f28/f32 retain their METAFOR-specific formulas with widened [-100,100] domains. This single-design protocol does not measure sampling uncertainty or independent noise repetitions. It is not interchangeable with the historical aggregates, 64-predictor native views or repeat pilot in v2.0.0, which remain available under their original DOI. Per-suite scaled tables are convenience outputs; cross-suite comparisons should fit a common feature-column scaler separately per dimension. No automatic performance join is supplied.