A laboratory Bell test with time-bin-entangled photons distributed through optical fibers reported a CHSH value of 2.583 with an uncertainty of 0.002, more than 265 standard deviations above the local-realistic limit, according to an arXiv preprint. CHSH is a statistic built from correlations across four pairs of measurement settings, with three correlations added and the fourth subtracted.
The experiment was designed around spacelike separation, random setting choices and checks for postselection effects. The paper says the detection loophole remained open, however, because the system did not reach the required overall detection efficiency and no-click events were discarded under a fair-sampling assumption.
The test was built around timing and distance
The source produced time-bin-entangled photon pairs, with the entanglement associated with different timing bins, and sent them to independent receiving stations known as Alice and Bob. The experiment used a fiber-based source for the time-bin entanglement.
Alice and Bob were separated by 49.0 metres, with a reported uncertainty of 0.7 metres, in a layout designed for spacelike separation. That term refers here to an arrangement intended to keep the relevant events outside each other’s light-speed reach during the measurement.
A quantum random-number generator, or QRNG, supplied four random bits for each Bell trial at 500 megabits per second, or 125 megahertz. The least significant bit selected each station’s binary measurement setting, providing the random choices used in the test.
Checks beyond the headline result
The detailed representative acquisition lasted about 1,650 seconds and accumulated around 1.9 million coincidence events. The paper says multiple independent Bell tests were performed and that their results varied by less than 5%.
The researchers compared marginal probabilities before and after coincidence filtering as a check related to no-signaling. The full-data estimate, made before filtering, was 0.0002 with an uncertainty of 0.0002 and was reported to contain zero within one standard deviation. The coincidence-conditioned estimate was 0.006 with an uncertainty of 0.001, almost six standard deviations above zero.
The experiment also analyzed all detected coincidence events without temporal filtering. The paper reports negligible impact on visibility and negligible side-peak detections for both receivers, findings used in its assessment of the postselection loophole.
Supplementary source characterization estimated the state fidelity at 0.9478, with an uncertainty of 0.0007, relative to the target entangled state. A numerical fit to an effective interference model reported an effective visibility of 0.92 and a fit statistic of 0.99969 for measured versus fitted conditional probabilities.
Randomness and timing were part of the evidence
The QRNG characterization reported per-round distance parameters of approximately 10 to the minus 49 and 10 to the minus 48 for Alice and Bob. Over the Bell-test duration, the corresponding values were approximately 10 to the minus 37 and 10 to the minus 36. The paper also reports that all tests for both QRNGs passed, with a 99% confidence interval for no technical issues from hardware or digital processing.
The event record listed a 172-nanosecond event for both parties, with an uncertainty of 1 nanosecond. It placed the detection event at 202 nanoseconds for Alice and 204 nanoseconds for Bob, each with an uncertainty of 2 nanoseconds.
The unresolved detector problem
The paper reports a critical overall detection-efficiency threshold of 0.828 for maximally entangled states using standard CHSH operators, but says the system did not reach that threshold. Because no-click events were omitted under a fair-sampling assumption, the result does not constitute a detection-loophole-free Bell test.
Taken together, the reported measurements support the paper’s treatment of locality, random setting choice and postselection, while leaving detection efficiency as a separate unresolved limitation. The coincidence-conditioned marginal difference also remains distinct from the full-data estimate, which was the estimate reported to include zero within one standard deviation.
Still a preprint
The document is an arXiv preprint, version v1, dated 28 August 2026. Its acknowledgments list support from EU Horizon Europe Marie Sklodowska Curie grants 956071 and 101072637, along with a John Templeton Foundation grant numbered 63132.
Paper data and sources
Original title: Postselection-loophole-free Bell test under strict spacetime constraints
Authors: Kannan Vijayadharan, Matías Rubén Bolaños, Andrea Pompermaier et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-28
DOI: Not available
Original paper · Full text