Imagine you're a wine critic blindfolded at a tasting. You sip from a glass, then sip again from the same glass a minute later. If the wine is distinctive enough — a bold Barolo, not watered-down table wine — you can tell the two sips are correlated in a way that no random pairing of wines could explain. But here's the trap: your ability to detect that correlation depends entirely on how distinctive the wine is, not on whether the glass was clean between sips. A dirty glass might ruin the actual wine, but your correlation test still passes because it was measuring the wine's character, not the glass's cleanliness. That structural mismatch — between what your test certifies and what the transmission channel actually preserves — is exactly the mechanism this paper dissects for quantum systems. The committed claim: for a single qudit sent through a noisy quantum channel, the temporal nonlocality robustness (TNR) — quantifying how strongly two-time correlations violate any classical explanation — is determined entirely by the input state's purity, not by the channel's coherence properties. TNR vanishes if and only if the input state is maximally mixed (ρA = 𝟙/d), and this holds across all standard noise families including complete decoherence. The resource is the state's distinguishability from randomness, full stop. This result cuts both ways. On the power side, TNR device-independently lower-bounds the fidelity of temporal teleportation — sending an unknown quantum state forward in time through a noisy channel. At dimension d=3, this bound reaches 7/9 without trusting the measurement devices. That's a genuinely useful certification tool. On the trap side, because TNR tracks state purity rather than channel quality, it can certify more than the channel actually delivers. The paper gives a sharp example: an injective unitary channel achieves the maximal temporal-Bell signal, yet its actual teleportation fidelity falls below the classical baseline. Your test says the channel is great; the channel is lying. The resolution is clean and complete. The authors prove a universal cap TNR ≤ (d−1)/d with an exact channel-resolved formula. For the depolarizing channel and for sufficiently mixed probes, honest certification holds — meaning the bound TNR provides actually tracks what the channel can do. But they also prove a no-go: no choice of probe states makes TNR certification universal across all channels. The over-certification problem is structural, not fixable by better experimental design. Underpinning everything is a unified semidefinite programming (SDP) hierarchy connecting three temporal quantum correlation measures: temporal entanglement robustness (TER), temporal steering robustness (TSR), and temporal nonlocality robustness (TNR). These form a strict lower hierarchy (TER ≥ TSR ≥ TNR) and an upper hierarchy conditioned on no-signaling in time (NSIT). The SDP framework makes the entire structure computationally tractable — all results are verified numerically for qudit dimensions d=2 through d=5. The paper sits in a fight that's been simmering since Leggett-Garg: does temporal nonlocality behave like spatial nonlocality, or is it fundamentally different? The spatial Bell story assigns nonlocality to the shared entangled state. This paper says temporal nonlocality is even more lopsided — it's entirely in the input, with the channel playing no role in the resource accounting. That's a conceptual advance, not just a technical one, because it means the certification guarantees you can extract from temporal correlations have a hard structural limit that spatial analogues don't share. The 21-page paper plus 12-page supplement is dense but internally consistent. The SDP verification across dimensions 2-5 is a solid computational check, though the absence of any experimental validation or independent numerical replication keeps this firmly in the theoretical-proof-plus-own-numerics category. The obvious next step — testing these bounds on an actual quantum processor with real noise — is not attempted, and the honest read is that this is a theory group without hardware access, not a result being hidden.
