Bell tests rest on a piece of electronics that rarely appears in the abstract: the coincidence monitor. It decides which detector clicks count as a single entangled pair. In the earliest optical Bell experiments, that monitor was not a commercial product. It was assembled by a nuclear physicist from surplus spectroscopy modules, and the choices baked into it still shape how replication attempts succeed or fail.

A Spare-Parts Anomaly at the Heart of Bell Tests

The logic of a Bell test is straightforward to state. Two detectors watch for photons from a shared source. Correlations between their outcomes, measured across different analyzer settings, test whether the source behaves in ways local hidden-variable theories forbid. The coincidence monitor is what makes each correlation number meaningful, because it filters out the vast majority of detector clicks that have nothing to do with the source. In the 1970s and 1980s, no vendor sold a coincidence unit tuned for optical Bell work. Nuclear physics labs already had the building blocks: fast discriminators, constant-fraction timing modules, and time-to-amplitude converters, all in the standardized NIM format. A physicist trained on nuclear spectroscopy could wire these into a coincidence circuit in an afternoon. That is essentially what happened, and the result was a device that lived in one lab and was never mass-produced.

The anomaly matters because the monitor sits between the physics and the published correlation coefficients. Any drift in its timing window, any miscalibrated dead-time correction, propagates directly into the measured violation of the Bell inequality. Historians of experiment have started treating that box as part of the apparatus, not as neutral plumbing.

How a Nuclear Physicist Built the Monitor

The build followed a pattern familiar from nuclear spectroscopy. Fast discriminators convert each photodetector pulse into a clean logic signal. A coincidence unit then asks whether two logic pulses arrived within a set window, typically a few nanoseconds. The physicist who assembled these monitors had spent years doing exactly this kind of work with scintillation counters, so the adaptation to optical detectors was incremental rather than radical.

Calibration was done with cosmic-ray muons, which provide a convenient source of genuinely simultaneous events across two detectors. A muon passing through both scintillators produces a coincidence that is physically guaranteed, so it can be used to set the timing offset between channels. The same trick transfers to optical setups with some care. Lab notebooks from the period show pages of muon runs, offset adjustments, and hand-drawn timing spectra.

Almost none of that made it into journal articles. Methods sections from the era describe the coincidence window in a sentence or two, if at all. The schematics, threshold settings, and calibration runs stayed in the notebook. That gap between what was done and what was published is the origin of most subsequent confusion.

The Procedural Choices That Shaped Results

The coincidence window is a first choice that matters. Set it too wide and accidental coincidences from unrelated detector clicks inflate the correlation rate. Set it too narrow and genuine pairs are discarded, which can bias the sample toward particular source configurations. A window of a few nanoseconds was typical, but the exact value depended on detector jitter and cable lengths, and it varied between labs.

Dead time is a second factor. Every detector and every logic module has a period after each event during which it cannot register another. If dead time is not corrected, high count rates suppress the coincidence rate in a way that mimics or masks a violation. The correction is arithmetic, but it depends on knowing the dead time precisely, and that number was often estimated rather than measured.

Threshold tuning was a third. Discriminator thresholds had to be set above electronic noise but below the single-photon signal. On the equipment of the day, that meant turning a screw and watching an oscilloscope. Two operators could reasonably land on different settings, and the effect on the final correlation was not always small. This is the kind of operator-dependent choice that later blinded analyses were designed to eliminate.

None of these choices were hidden deliberately. They were simply not considered part of the result. The physics was the correlation curve; the electronics were assumed to be transparent. That assumption is exactly what replication audits now question.

Why the Build Details Matter for Replication

Later Bell tests used different coincidence electronics, often commercial time-to-digital converters or field-programmable gate arrays. Each generation improved timing resolution and reduced dead time. But improved hardware does not automatically reproduce earlier results, because the earlier results were produced under the earlier hardware's specific biases. Comparing across generations requires knowing what those biases were.

Small timing offsets can mimic violations. A systematic delay between two channels that shifts with analyzer setting produces a correlation pattern that looks like entanglement but is not. This is not a hypothetical concern. Timing-related artifacts have been identified in several subfields where coincidence logic is central, including the stellar occultation timing discrepancies documented in a 2019 occultation timing log.

The original schematics for early Bell-test coincidence monitors are incomplete or lost. Some survive as hand-drawn circuit diagrams in institutional archives. Others are known only through the memory of the people who built them. Replication attempts that omit detector dead time, or that use a coincidence window an order of magnitude different from the original, are not strictly replications at all.

Historical audits have turned up undocumented calibration steps in adjacent fields. The pattern is consistent: the published result is reproducible in spirit, but the numerical agreement depends on choices that were never written down. A 2016 reanalysis that reversed a pain study's effect after its blinding log mismatched, described in the reanalysis account, is a useful parallel from outside physics.

Reproducibility Trade-offs in Modern Entanglement Experiments

Custom hardware belongs in the peer-reviewed record. A schematic in a supplementary file costs little and prevents decades of ambiguity. Some modern groups already do this, particularly those working on loophole-free Bell tests where every timing detail is scrutinized. Others still describe their coincidence logic in a paragraph.

Calibration data should travel with results. If the coincidence window was set using muon runs, those runs should be published alongside the correlation curve. The data is not glamorous, but it is the difference between a result that can be checked and one that can only be admired.

Open-source coincidence designs are emerging as a partial fix. A shared reference design, with published timing characteristics and dead-time behavior, gives independent groups a common baseline. It does not eliminate hardware-specific effects, but it makes them comparable. The trade-off is that a reference design can also become a shared blind spot, if its assumptions are wrong in the same way everywhere.

Treating timing electronics as part of the apparatus, rather than as infrastructure, changes how experiments are reported. It also changes how they are reviewed. A referee who understands the coincidence monitor will ask different questions than one who treats it as a black box.

Documenting Timing Electronics

Archive oscilloscope traces and threshold settings for every run, not just the ones that produced publishable data. The discarded runs are often the most informative about systematic behavior.

Report the coincidence window and the dead time explicitly in the methods section, with units and the method used to determine each. A range is acceptable; silence is not.

Cross-check timing with an independent module from a different manufacturer. Agreement between two dissimilar systems is stronger evidence than repeated measurements on one.

Include hardware schematics in supplementary materials, even when the circuit is commercially available. The specific revision and configuration matter.

Run a blind analysis on timing parameters, where the person choosing the coincidence window does not see the correlation curve until the choice is locked. This is the same logic that motivated blinding in clinical trials, and it applies here for the same reason.

None of this guarantees that a given Bell test is correct. It does make the reasoning inspectable, which is the most that any experimental protocol can offer.