Imagine you're a bank auditor, but you're not allowed inside the vault. You can only slide questions under the door and read the answers that come back. Your job is to figure out — from the pattern of answers alone — whether the vault actually contains real gold or just painted bricks. That's self-testing in quantum mechanics: you run a statistical game (the CHSH test) and the correlations in the outputs certify what quantum state was inside, without ever directly measuring it. The catch has always been that the theory assumed perfect gold bars — ideal GKP states with infinite energy — while real labs produce slightly tarnished ones. Salek and Hayashi's committed claim: you can run a CHSH Bell test on realistic, finite-energy GKP states in a full bosonic oscillator and still extract a device-independent certificate that you have a genuine Bell pair. Not a logical-qubit approximation. Not a post-selected subset. The physical correlations themselves, computed across all photon-number sectors, do the work. The mechanism is a clever measurement design. Position and momentum bins give you the standard Pauli X/Z settings. The tilted CHSH settings — the ones that actually distinguish quantum from classical — come from coarse-graining photon number modulo four and a displaced conjugate. With the canonical displacement d = √π, the CHSH score exceeds the classical bound of 2 once per-peak squeezing passes 4.56 dB, and Kaniewski's extractability bound (the one that actually certifies a Bell pair) kicks in above 5.02 dB. These are not heroic numbers — current optical squeezing experiments routinely hit 12-15 dB. The real leverage comes from a single calibration step. By tuning the displacement parameter d using an independently measured finite-energy parameter — crucially, fixed before any Bell-test data are collected — the thresholds drop to 4.21 dB and 4.58 dB respectively. At 12 dB squeezing, this calibration lifts the CHSH score from 2.695 to 2.789 and the target-state overlap bound from 0.908 to 0.972. That last number matters: 0.972 means the extracted state is within 3% of a perfect Bell pair. The paper is careful to note this gain is specific to deterministic phase-bit coarse-graining; randomizing the odd-sector outcomes gives only marginal improvement. On the integrity front, this is an analytical/theoretical paper — no experiment, no simulation. The validation is mathematical proof: closed-form computation of correlations for number-filtered GKP states. The authors are explicit that their thresholds are honest-model predictions, not accounting for optical loss, detection efficiency, or finite-sample statistics. That's an important caveat. A real experiment would face all three simultaneously, and each one erodes the CHSH score. The paper essentially says: here's the idealized-but-finite-energy floor. Real experiments start here and get harder. The field fight this lands in is whether continuous-variable (bosonic) quantum systems can achieve the same device-independent certification guarantees that discrete qubit platforms already demonstrate. Prior CHSH self-testing results either assumed infinite-energy ideal GKP states or worked only in the qubit logical subspace with post-corrections. This paper argues that you can stay in the full infinite-dimensional Hilbert space of the oscillator and still get robust self-testing, which is the prerequisite for device-independent quantum key distribution over optical channels. The obvious next experiment nobody ran: an actual optical Bell test with GKP states at ≥5 dB squeezing, including loss and detector-efficiency modeling. The honest read is (a) — this is a theory group (Salek and Hayashi are information theorists), and building the experiment requires a very different lab. The paper is a proof-of-principle invitation to experimentalists, not a claim that self-testing has been demonstrated. The gap between 0.972 overlap in theory and a loophole-free Bell test with GKP states in a real fiber channel is where the next 3-5 years of work lives.
