Imagine you're handed a locked piano. You can't open the lid to see which strings are inside, and someone has rewired every key to hit a random string. But you can press keys and listen. If you're clever about which keys you press and how you analyze the sound, you can reconstruct the tuning of every individual string — without ever opening the lid. That's the core mechanism of this paper. The committed claim: given a Hamiltonian whose non-interacting quasiparticle structure is scrambled by an unknown global unitary U, the authors' algorithm SPICES (Single-Particle Inference via Contour Estimators) can efficiently recover the single-particle energy spectrum {Ei} in Wasserstein distance, without reconstructing U itself. The paper also proves this problem is DQC1-hard — meaning even though the Hamiltonian decomposes into non-interacting modes, learning the energies is at least as hard as the hardest problems solvable by the "one clean qubit" model of computation. The non-interacting structure does not make the problem easy. The architecture is elegant in its indirection. SPICES uses the Hadamard test — a standard quantum subroutine that estimates expectation values of unitaries — to sample the partition function Z(β) = Tr(e^{-βH}) at complex inverse temperatures β. The key mathematical insight is that zeros of Z(β) in the complex plane encode the single-particle energies as Fourier frequencies. The classical postprocessing then becomes a spectral estimation problem: recover frequencies from sampled values of a quasi-periodic function. The authors develop a contour-based estimator that avoids the usual pitfalls of superresolution (ill-conditioning, noise amplification) by exploiting the analytic structure of partition functions. This is a non-variational approach — no optimization loop, no barren plateaus, no ansatz design. It sits in the Hadamard-test family alongside randomized benchmarking and DQC1 trace estimation, not the variational quantum eigensolver (VQE) family. The ladder comparison is where honesty matters. This is not a paper that claims to beat classical DFT or CCSD(T) on any existing benchmark molecule. It introduces a new problem formulation — Hamiltonian learning under obfuscation — that didn't have a prior SOTA. The relevant classical comparators are mean-field theory and Green's function methods, which assume you already know the basis (i.e., U is identity). The paper's contribution is addressing the case where you don't know U, which is arguably the physically relevant scenario for strongly correlated materials where the natural quasiparticle basis is unknown. The DQC1-hardness proof establishes that no efficient classical algorithm can solve this in general, which is a meaningful complexity-theoretic statement. Integrity is solid for a theory paper. The DQC1-hardness result is a formal proof, not a simulation. The SPICES algorithm comes with explicit sample complexity bounds and convergence guarantees in Wasserstein distance under stated assumptions (bounded spectator norm, gap conditions on the single-particle spectrum). There are no cherry-picked benchmarks because there are no numerical benchmarks at all — this is a 70-page theory paper with 3 figures. The assumptions are clearly stated, which is the right kind of honesty for this stage of work. The absence of any numerical demonstration is both a strength (no cherry-picking possible) and a limitation (no empirical validation of constants). The milestone gap is real. SPICES is a quantum algorithm that requires coherent evolution under e^{-iHt} for times scaling with the inverse spectral gap. Running it on actual quantum hardware for a materials-relevant Hamiltonian (say, 50+ orbitals of a transition metal oxide) requires fault-tolerant quantum computers that don't exist yet. The path from here to application runs through the same bottleneck as every other quantum simulation algorithm: logical qubit counts, gate depths, and error correction overhead. The paper's value today is as a complexity result and algorithmic primitive, not as a near-term tool. The obvious next experiment the authors did not run is a numerical simulation of SPICES on a small system (say, 8-12 qubits) with realistic noise models. At 70 pages of theory, this reads like a deliberate scoping decision — the authors chose to establish the full theoretical foundation first. The numerical companion paper is almost certainly in preparation or being saved for a separate submission. A second gap: the paper assumes access to time evolution e^{-iHt}, which itself requires a Hamiltonian simulation subroutine. The end-to-end resource accounting (total gates, total shots) for a concrete material is not provided.
