Uncorrected relative peripheral refractive error and its biometric determinants in a GRIN ray-tracing pediatric eye

Model-output data for the bare (uncorrected) eye — model validation, clinical consistency, and biometry-sensitivity / hypothesis generation — accompanying the manuscript on peripheral defocus and myopia. Generated by the open grin-defocus-eye engine v1.1.0 (software record DOI 10.5281/zenodo.20739278). Each row is a direct engine readout — no post-processing. Model. AVOCADO gradient-index crystalline lens (Sheil & Goncharov) in an aspheric pediatric eye (age 10); own conic tracer and RK4 GRIN integrator (a homogeneous Le Grand eye on Optiland is a validation oracle only). The retina is an Atchison conicoid coupled to axial elongation (not to the central refraction): axial ametropia reshapes the retina, refractive ametropia keeps the emmetropic retina — so the axial-vs-refractive comparison is a clean optics-vs-globe decomposition. Accommodation is the AVOCADO volume-constant reshape with in-vivo Dubbelman rates recalibrated (×1.583) to a ~1:1 central gain. Conventions. Dioptres (D); sign clinical (myopia < 0, hyperopia > 0); RPRE(θ) = M(θ) − M(0) (relative peripheral refractive error; >0 = relative peripheral hyperopia, myopiagenic); J0 > 0 = with-the-rule astigmatism. Field angle = retinal eccentricity. Two viewing states: distance (relaxed, A=0, cycloplegic) and near33 (33 cm, accommodated). Grid: pupils 2/4/6 mm, eccentricity 10/20/30°. Peripheral readout validated against the homogeneous oracle to ≤ 0.03 D out to ~20°; 30° reported with a caveat; 40° out of trust. License: CC–BY–4.0. Files. data/bareeye_ametropia.csv — uncorrected relative peripheral refraction (and J0) across the full ametropia series (SR −6…+6 D), distance and near. Reproduces the clinical pattern: myope → relative peripheral hyperopia (−6: +0.60 D @30°), emmetrope −0.46 D @30° (vs Mutti 2000 −0.41), hyperope +3 −1.01 D. Columns: SR, eye_M0, cond, A_acc, pupil, angle, M0, RPRE, J0. data/biometry_grid.csv — tidy long-format biometry-sensitivity grid across the full refraction series. One row per (refraction, block, factor level, condition, pupil, angle). Columns: R, block, factor, factor_value, cond, A_acc, pupil, angle, M0, RPRE. Blocks: H1 (iso-refraction axial length, cornea refit to hold central refraction; factor_value = AL mm), H3_cornea_ecc (anterior corneal asphericity as eccentricity, Q=−e²; factor_value = e), ACD (lens position ±0.3 mm at fixed globe, cornea refit; factor_value = dACD mm), axial_vs_refractive (factor_value = axial or refractive). data/biometry_grid.json — the same grid in the raw nested structure (per refraction → block → factor level → state → pupil → angle), including solved geometry (AL, Rcf, K) and accommodation per cell. Key results encoded in the data. Clinical consistency (bare eye, distance): monotonic transition from relative peripheral hyperopia in myopes through slight peripheral myopia in the emmetrope to relative peripheral myopia in hyperopes, matching Mutti 2000 at the anchors. Axial vs refractive: axial ametropia carries ~2–3× the peripheral defocus of refractive ametropia of equal power — about half to two-thirds of a myope’s peripheral hyperopia is retinal (globe shape), not purely optical. Corneal eccentricity: a refraction-independent additive lever, +0.224 D per +0.1 eccentricity at 30°, constant across all 13 refractions; high eccentricity can flip an otherwise protective myopic periphery to myopiagenic hyperopia. Anterior chamber depth: lens forward (deeper aqueous) → peripheral hyperopia, ~+0.23 D per +0.15 mm. Accommodation: demand-driven peripheral hyperopic shift ~0.13 D per dioptre of accommodation, nearly independent of corneal eccentricity. Provenance. Numpy-only, deterministic. Generating scripts (in the software record): experiments/run_part1_grin.py, experiments/grin_part1.py, experiments/run_part1_fullgrid.py, experiments/part1_grid_to_csv.py. Please cite both this dataset and the software record (DOI 10.5281/zenodo.20739278). See DATA_README.md for the full data dictionary.