If glitches cluster non-randomly in time, could detector networks eventually become *less* discriminant as more detectors add correlated noise?

You've identified a genuinely worrying scenario. Let me walk through what the archive says about this.

The core tension is real: as detector networks expand, they should gain statistical power to distinguish gravitational-wave signals from noise. But if glitches don't arrive randomly—if they cluster in time or correlate across detectors—then simply adding more detectors could paradoxically flood the analysis with correlated junk.

The clustering problem is confirmed in the data. Several glitch morphologies violate Poissonian statistics. Specifically, glitches classified as "1400Ripples," "Fast Scattering," and "Power Line" show significant departures from random arrival times [AG-2025.11-155], suggesting they cluster or have memory. This directly challenges the assumption that more detectors = more independent noise witnesses.

The false-alarm problem is already observed. A recent study quantified detection confidence in multi-detector networks using the BayesWave burst-search pipeline. The key finding: while the signal-versus-glitch Bayes factor should improve with detector count, the *frequency* of non-Gaussian transients also increases with network size. When glitches occur more often, false alarms rise, and overall detection efficiency can degrade despite better theoretical discrimination [AG-2024.03-419]. This is your scenario playing out empirically.

Why it matters: if detector networks become saturated with correlated glitches, you hit a point of diminishing returns—more data doesn't help if it's all contaminated in the same way.

Some defenses exist, but they're labor-intensive. Researchers can mitigate overlapping glitches by excluding marginal cases or by using sophisticated Bayesian methods that model glitch morphology [AG-2026.04-211]. More recent work on singular-value-decomposition discriminators shows promise in distinguishing glitch "fingerprints" from signals [AG-2025.07-142], and automated time-frequency localization of glitches is being developed to replace manual vetting [AG-2025.11-470]. But these are all post-hoc patches.

The real worry you've raised: if glitch clustering is intrinsic to detector physics (not random), then the fundamental assumption underlying multi-detector inference—that independent detectors provide independent noise samples—breaks down. The archive doesn't directly address whether or how to recover discriminant power once correlation dominates.