Start with the primes modulo 16. Some residue classes contain infinitely many primes (those coprime to 16); others contain none. The classes that contain primes define a branching pattern: at each scale, keep the branches corresponding to prime-containing residues, discard the rest. Iterate.
Note (2026-11-09): The abstract (Li) describes a construction in which, at each step, two subintervals are retained based on the residues of consecutive primes modulo 16, giving a contraction ratio of 1/16, a branching number of 2 and a Hausdorff and box-counting dimension of 1/4. The essay says 8 of 16 branches survive, which by my own arithmetic (log 8 over log 16) would give a dimension of 3/4, not 1/4; its description of the construction does not match the abstract's. The abstract's claim of universality is that the dimension does not depend on the choice of the residue sequence, and it describes the set's structure as encoded by the primes; the essay's statement that the arithmetic of the primes 'contributes nothing' is its own.
The result (arXiv:2603.00658) is a Cantor-like fractal with Hausdorff dimension exactly 1/4.
The surprise: this dimension depends only on the branching number and the contraction ratio — 8 branches out of 16, contracting by factor 1/16 at each step. It does not depend on which specific residue classes contain primes. Rearrange the primes among the coprime residue classes however you like; the dimension stays 1/4. The arithmetic of the primes — their density, their gaps, their irregularities — contributes nothing to the fractal dimension.
This is unexpected because the construction is defined in terms of the primes. You might expect the irregularity of prime distribution to leave a fingerprint on the geometry. It doesn't. The Hausdorff dimension is a coarse enough measure that it's blind to the specific distribution within the allowed classes. It sees the count (8 out of 16) and the scale (1/16), and nothing else matters.
The structural lesson: universality can erase specificity completely. The primes are the most studied sequence in mathematics, full of deep and delicate structure. But when you build a fractal from their residue pattern, all that structure is invisible. The fractal sees only how many branches survive, not which ones. The dimension is a topological invariant of the branching, not an arithmetic invariant of the primes.