Imagine you're sorting a shuffled deck of cards. You peek at a few cards from the top, hoping to learn something about the rest. If the deck is shuffled well enough, those peeks tell you almost nothing — the remaining cards are still distributed as if you'd never looked. Deep thermalization is the quantum version of this: measure a big chunk of a quantum system, and the leftover unmeasured subsystem collapses into a pure state drawn from a universal distribution, independent of the microscopic details you revealed. This paper asks: does that universality hold for free fermions — quantum particles that don't interact — and can we write down the distribution explicitly? The committed claim: in a free-fermion chain subject to Gaussian-preserving local measurements, the projected ensemble of the unmeasured subsystem converges to a fermionic Gaussian Scrooge ensemble determined by the subsystem's generalized Gibbs ensemble (GGE). This is not a small refinement. Previous deep-thermalization results focused on chaotic (interacting) systems where eigenstate thermalization does the heavy lifting. Integrable systems like free fermions have infinitely many conserved charges that prevent standard thermalization — yet the authors argue universality still emerges, just governed by GGE rather than a thermal Gibbs state. The architecture sits squarely in the free-fermion / Gaussian-state toolkit: correlation-matrix methods, Wick's theorem, and Grassmann integration. The measurements are local and break particle-number conservation while preserving Gaussianity — a deliberate technical choice that keeps the state manifold tractable. The key analytic move is showing that measurement outcomes carry information about local conserved charges in a finite subsystem, but that this information becomes negligible in the thermodynamic limit. They call this 'effective non-revelation,' and it is the engine that drives universality. The ladder here is unusual because this is not a computational-performance claim. The relevant prior art is the Scrooge ensemble framework of Goldstein et al. and the deep-thermalization program of Cotler et al. (chaotic systems) and Ho & Choi. For interacting systems, deep thermalization to Haar-random projected ensembles is established; for integrable systems, only reduced-density-matrix level thermalization (to GGE) was known. This paper extends the description one level deeper — from the density matrix to the full statistics of conditional pure states. That is a genuine conceptual advance within integrable-system thermalization, not an incremental metric improvement. Integrity is mixed in the way typical of mathematical-physics papers: the core argument is analytical, supported by numerical checks on finite systems. The authors construct the Gaussian Scrooge ensemble directly at the subsystem level and then verify numerically that the late-time projected ensemble at finite temperature matches. The numerics are finite-size (as they must be for exact diagonalization of correlation matrices), and no independent replication exists yet. The paper is 28 pages with 3 figures — this is a theory paper earning its keep through derivations, not benchmarks. The milestone framing in this subfield is not about beating a number — it is about extending the deep-thermalization program to the next class of systems. Free fermions are the simplest integrable systems; the obvious next targets are interacting integrable models (Bethe-ansatz solvable chains like XXZ) and then many-body localized systems where thermalization is believed to fail entirely. Whether the Scrooge-ensemble picture survives interactions and disorder is the load-bearing question for the next 3-5 years. The experiment the authors did not run: interacting integrable systems. The Gaussian machinery that makes free fermions tractable breaks down entirely when you add interactions. This is almost certainly a case of (a) — the method doesn't extend, not that they're hiding negative results. The honest next step requires genuinely new tools, likely tensor-network or Bethe-ansatz-based constructions of projected ensembles, which nobody has yet.
