This work falls within the scope of recreational mathematics and the experimental exploration of the order-3 magic square of squares. Instead of resolving the open problem of the existence of such an object, the research focuses on "near-miss" structures. The analysis is based on the reduction of the original problem to three three-term arithmetic progressions of square numbers with a common difference (r), whose first terms also form an arithmetic progression.By relaxing the condition requiring the initial terms to form an arithmetic progression, this paper introduces a percentage error function (p). This function allows for a quantitative assessment of the discrepancy between the differences of these terms, and consequently, a quantitative measure of how closely a configuration approximates a fully magic structure.Subsequently, an algorithm developed for this study was applied to generate all three-term progressions of integer squares with a common difference r. To optimize calculations and eliminate redundant operations, a filter based on the theory of quadratic residues in the ring Z_240 was implemented. Ultimately, a deterministic exploration of the differences r from 2 to 70,000,000,000,000 was conducted.Next, a geometric parametrization utilizing 120-degree Eisenstein triangles is presented. This enabled a dimensionality reduction of the error optimization problem to a function of a single real variable. It was demonstrated that the theoretical optimum point with zero error corresponds to an irrational number, preventing its direct realization in integers. It was proven, however, that despite structural limitations, it is possible to construct a parametric matrix generating semi-magic squares with equal sums in rows, columns, and on one diagonal. Diophantine approximation and continued fractions yielded an infinite sequence of rational approximations where the error function converges asymptotically to zero, while the ratio of the diagonal sums remains close to 1. This is illustrated by a numerical example where the ratio yields a value on the order of 1 - 7.3506 x 10^(-103230).The results offer a new method for the assessment of near-miss structures based on the error function p. They outline an approach based on Eisenstein triangles, for generating configurations where the error p tends to zero and the diagonal sum ratio approaches 1. This work does not prove the non-existence of the magic square of squares, but provides a universal research framework for approaching ideal structures.