Imagine you're trying to hear a faint radio station in a noisy room. One antenna picks up some signal, but the noise floor is brutal. Now imagine you could clone that antenna — not amplify the signal, but create a second independent copy of the listening channel, so that any noise that washes out Channel A might not wash out Channel B. You haven't boosted the transmitter; you've given yourself more chances to catch the whisper. That's the core mechanism here: repeated Stern-Gerlach-style splittings of a massive particle's spatial wavefunction create additional interferometric branches, each of which accumulates its own gravitational phase. The paper proves that these extra branches can only increase (or at worst preserve) the entanglement negativity — they never subtract from it. The committed claim: for two massive particles in spatial superposition, performing n splitting stages instead of one produces a density matrix whose entanglement is provably at least as large as the (n−1)-split case. The authors derive this via a local parity dephasing relation linking the n-split state to the (n−1)-split state, showing that two of the resulting four sectors reproduce the previous state while the other two pick up additional gravitational phases. An exact lower bound on the change in negativity follows from the negative partial-transpose eigenspace of the preceding state. The ladder here is sparse. The foundational single-split proposals — Bose et al. (2017) and Marletto & Vedral (2017) — are the baselines. Those protocols use one splitting per particle, generating a 2×2 grid of interferometric paths. This paper generalizes to 2^n paths per particle. There is no competing repeated-splitting proposal to compare against; this is more "first formalization" than "beats SOTA." The advantage is shown numerically in positional negativity over the full range of decoherence rates considered, but the numbers are theoretical — no experimental comparison exists. Architecturally, this lives in the family of Stern-Gerlach interferometry proposals for tabletop quantum gravity tests. The method leans on controlled unitary splittings of spatial degrees of freedom, with gravitational Newtonian phases accumulated between branches of two spatially separated masses. The key structural insight is the use of a reusable two-level internal system plus local classical memory registers to preserve full branch-label information through multiple splits. Without the memory register, a restricted readout is analyzed via a 2×2 Schur complement condition. Integrity is mixed. The results are a combination of analytical proofs (the monotonicity bound on negativity, the exact Schur complement condition) and numerical simulations. The analytical results are self-contained and appear to be exact — not approximate. The numerics are confirmatory, not the primary evidence. However, this is entirely theoretical: no experiment, no independent simulation, no pre-registration. The authors are honest about the gap between protocol and implementation — they note that the required coherence times and splitting fidelities are at the frontier of what's currently achievable. The milestone question is really about the experimental programs pursuing the single-split version. Groups at UCL, Warwick, and elsewhere are working toward the first single-split gravitational entanglement detection, which requires masses of ~10^{-14} kg held in superposition for ~1 second over ~100 μm separations. The repeated-splitting protocol adds complexity (more splitting stages, memory registers) but relaxes the per-shot entanglement requirement. The next concrete number: if a single-split experiment achieves marginal but inconclusive entanglement (say, negativity barely above noise), this protocol gives a rigorous path to boost the signal without changing the mass or separation. The obvious next experiment the authors didn't run: a full decoherence model with realistic noise channels (not just dephasing but motional heating, splitting-pulse imperfections, gravitational gradient noise) applied to the repeated-splitting protocol, benchmarked against the single-split case under identical noise budgets. The honest read is (a) — this is a theory paper establishing the mathematical framework, and the realistic noise modeling is a substantial computational and modeling effort that belongs in a follow-up or collaboration with an experimental group.
