Imagine you're packing a suitcase with rulers, and each ruler has to be oriented so it tells you absolutely nothing about the readings on any other ruler. In three dimensions you can fit four such rulers (mutually unbiased bases), and in any prime-power dimension d the answer is known: d+1. But in composite dimensions — 6, 10, 12, 36 — nobody knows the maximum, and the best anyone could do was bolt together smaller solutions using tensor products. This paper finds a way to squeeze in more rulers than that bolt-together method allows. The committed claim: a new ansatz — diagonal phase matrices applied to tensor products of Fourier and real Hadamard matrices — constructs more MUBs than previously known in multiple composite dimensions, and the count grows as √d using Paley Hadamard matrices, beating the tensor product bound asymptotically in every eighteenth dimension. This is not incremental. The MUB existence problem in non-prime-power dimensions has been stuck for decades, and prior constructions topped out at the tensor product ceiling. The construction is elegant in its modularity. You pick a fixed unitary that's a tensor product of discrete Fourier transforms (and optionally a real Hadamard matrix), then generate each basis by conjugating with a diagonal phase matrix. The key insight is that mutual unbiasedness reduces to checking inner products between columns of these phase matrices — a condition the authors translate into combinatorial constraints they can solve explicitly. In dimension 12 they get 5 MUBs (previous best: 4 from tensor products). In dimension 36, they get 7. In dimension 648, they get 10. The Paley construction gives q+1 MUBs in dimension q(q+1) for every prime power q ≡ 3 mod 4, and that count scales as √d — provably unreachable by any tensor product decomposition. The ladder here is the tensor product bound itself, which has been the practical ceiling for decades. In dimension 12, tensor products give 4 MUBs; this paper gives 5. In dimension 648, the tensor product bound yields fewer than 10; this paper hits 10. The theoretical maximum in dimension d is d+1 (achievable only in prime-power dimensions), so in d=12 the ceiling is 13 — meaning 5 is still far from the theoretical limit, but it's the first time anyone has broken past 4. The real headline is the asymptotic growth: √d vs. the tensor product bound which grows more slowly in the relevant dimension families. Integrity is strong for a math paper: the results are constructive proofs, not numerical experiments. Each claimed MUB set can in principle be verified by checking the defining inner-product conditions. The paper provides explicit constructions and counts, not heuristic searches. The Paley-based result comes with a formal proof that the count exceeds the tensor product bound. There's no simulation to question — this is theorem-based mathematics. The main caveat is that the ansatz doesn't claim to be optimal; it's a lower bound improvement, not a proof that the true maximum has been found. The milestone everyone in quantum information theory watches is the MUB problem in dimension 6: does a complete set of 7 MUBs exist? This paper doesn't crack dimension 6 (the ansatz requires at least one tensor factor to be a Fourier matrix of prime-power size, and 6 = 2 × 3 doesn't yield improvement over the known 3 MUBs). But it reframes the problem: if diagonal-phase constructions can beat tensor products in d=12, 36, 648, the question becomes whether a more general ansatz — perhaps with non-diagonal entangling phases — could push the count in d=6 past 3. The obvious next experiment the authors didn't run: extending the ansatz to allow entangling (non-diagonal) phase matrices, or searching computationally for MUBs in dimension 6 using the structural insights from this work. The honest read is (c) — they're saving the dimension-6 attack for a follow-up. The current ansatz provably cannot improve d=6, so they'd need a genuinely different construction, and that's a separate paper's worth of work.
