Burstiness – the pattern of intense activity clusters followed by quiet periods – appears across nature and society, from gene expression firing in bursts inside cells to stock trades flooding exchanges before long lulls. For over 15 years complexity scientists have relied on the canonical Burstiness Parameter (BP), introduced by Goh and Barabási, which measures how much inter-event timing deviates from a perfect exponential (Poisson) process. A new preprint argues this workhorse statistic produces too many false negatives, quietly misclassifying moderately bursty power-law distributions as non-bursty under realistic conditions.

Posted to arXiv in April 2026 (identifier 2604.05188), the solo-authored preprint by Joshua Stadlan is not yet peer-reviewed. The author first proves analytically that BP fails for certain exponents of power-law and truncated power-law distributions. He then introduces the Burstiness Tail-based Index (BTI), a quantile-ratio metric that compares differences between upper and lower quantiles of the waiting-time distribution. This preserves the spirit of BP – comparing real data against a memoryless null – while correctly flagging burstiness where BP does not.

Methodology combined three approaches: mathematical derivation across parameter spaces, Monte Carlo simulations drawing repeated samples from known bursty distributions (testing sample sizes from hundreds to tens of thousands of events and varying observation windows), and a re-analysis of a previously published human communication dataset. The simulations demonstrated BTI remains stable under limited sampling and short windows where BP degrades. The empirical case study showed that switching from BP to BTI changes the inferred characteristic timescales of bursts in the human data, suggesting earlier work may have misjudged whether bursts operate on minutes, hours, or days.

This matters beyond methodology. Barabási’s landmark 2005 Nature paper ("The origin of bursts and heavy tails in human dynamics") used BP-style thinking to overturn the assumption that human activity follows Poisson statistics. Subsequent studies applied the same lens to email, mobile calls, Wikipedia edits, neuronal spiking, financial order flow, and ecological foraging. Yet if BP systematically under-reports burstiness in moderate regimes, many of those conclusions understated the strength of underlying feedback loops and memory effects. In finance, volatility clustering might be even more pronounced than BP analyses suggest; in molecular biology, transcriptional bursting – already known to drive cell-to-cell variability – could be more extreme, affecting models of gene regulation.

What most coverage of burstiness research has missed is the uncomfortable gap between theoretical power laws and practical detectability. Real datasets are finite, noisy, and censored by observation windows. Stadlan’s simulations quantify exactly how quickly BP’s false-negative rate climbs as sample size drops – a practical concern rarely discussed when researchers cite a single BP value from a few weeks of data. BTI’s robustness here is its strongest practical advantage.

Synthesizing these threads with a 2008 EPL paper by Goh and Barabási on "Burstiness and memory in complex systems" reveals a deeper pattern: the field has treated burstiness and memory as separate but often co-occurring features. A better detector like BTI lets researchers disentangle them more cleanly, potentially improving everything from epidemic models on bursty contact networks to algorithms that forecast intermittent traffic in communication systems.

Limitations remain. The preprint examines specific families of heavy-tailed distributions; performance on multimodal or non-stationary real-world signals needs further testing. As pre-peer-review work, claims about "dramatically reduced false negatives" await independent validation. Sample sizes in the human case study inherit whatever constraints the original dataset carried, and the exact simulation counts are only summarized.

Even so, the quantile-ratio lens offers a quietly radical shift. By focusing on tail quantiles rather than variance-to-mean ratios, BTI aligns better with how extreme events dominate complex systems – a insight that echoes everything from turbulence intermittency to earthquake aftershock statistics. For time-series modeling across biology, finance, infrastructure, and social systems, adopting BTI could mean fewer hidden mechanisms left undetected and more accurate forecasts of when the next burst will arrive.