On the exceptionality of exceptional gravitational-wave events Rodrigo Tenorio, Davide Gerosa Abstract In gravitational-wave astronomy, as in other scientific disciplines, “exceptional” sources attract considerable interest because they challenge our current understanding of the underlying (astro)physical processes. Crucially, “exceptionality” is defined only relative to the rest of the detected population. For instance, among all gravitational-wave events detected so far, GW231123 is the binary black hole with the largest total mass, while GW241110 is the binary black hole with the most strongly misaligned spin relative to the orbital angular momentum. Mandel 2026 Mandel 2026 argued that apparent “exceptionality” may reflect measurement error rather than an extreme true value, and suggested that the total mass of GW231123 may be significantly overestimated. Here we present a quantitative analysis that supports this conceptual point. We find that claims of “exceptionality” obtained under agnostic priors should be critically questioned whenever measurement uncertainties are comparable to the width of the underlying population. Specifically, we find that the total mass of GW231123 is unlikely to be meaningfully affected by this effect while the spin of GW241110 is far less likely to be anti-aligned than initially claimed: about 70% of realizations that appear to yield an “exceptionally anti-aligned” spin are in fact consistent with either nonspinning or aligned configurations. Introduction — Gravitational-wave (GW) detection of black-hole (BH) binaries is now becoming routine. In this context, it is natural to single out some detected events as “exceptional” because of their informative properties, distinguishing them from the bulk of the population (and, in practice for researchers, motivating dedicated “single-event papers”). Most often, such exceptional events correspond to extreme values in some of the binary parameters (masses, spins, redshift) compared to the other events observed so far. For example, GW231123 Abac et al. 2025a is the most massive GW event detected at the time of writing, with a total mass of M_{\mathrm{tot}}=238^{+28}_{-49},\mathrm{M}_{\odot} (90% credibility), and has sparked considerable interest because of its implications for, e.g., pair-instability processes in supernovae and hierarchical BH mergers. The previous record holder for the largest total mass from the preceding data-taking period, GW190521 Abbott et al. 2020 , sparked similar interest at the time. Events GW241011 and GW241110 Abac et al. 2025b present the highest spin component aligned/anti-aligned with the binary orbital angular momentum, respectively \chi_{1\mathrm{z}}=0.66^{+0.08}_{-0.09} and \chi_{1\mathrm{z}}=-0.39^{+0.34}_{-0.37} (90% credibility), with important consequences for the rate of binary BHs in dynamical formation channels. These measurements are obtained by conducting Bayesian inference on the observed data d using a waveform model parametrized by parameters \theta . Specifically, the quantiles quoted above are summary statistics of the posterior distribution p(\theta|d,\mathcal{I})\propto p(d|\theta,\mathcal{I})p(\theta|\mathcal{I}) , where the likelihood p(d|\theta,\mathcal{I}) encapsulates the generative process of the data (including noise model, detector properties, signal waveform model, etc.) and the prior p(\theta|\mathcal{I}) encapsulates our initial belief on the observed signal parameters. Note that inference is conditioned on the broader context \mathcal{I} , which is typically omitted from the notation and collects any unstated but relevant assumptions included in the inference process Jaynes 2003 . Mandel 2026 Mandel 2026 recently highlighted that exceptional events are so because their inferred parameters are exceptional compared to a population. The emphasis on the word inferred is important here: such inference (a.k.a. parameter estimation) is typically performed using agnostic assumptions. Let us denote this context by \mathcal{I}_{\mathrm{ag}} ; GW practitioners commonly refer to this as “the PE prior.” Crucially, \mathcal{I}_{\mathrm{ag}} ignores the rest of the population. We therefore find ourselves in a conundrum: an event is deemed “exceptional” relative to a population, yet the “exceptionality” itself ignores the population. Such statements of exceptionality are therefore ill posed because the information required demands a different set of assumptions, \mathcal{I}_{\mathrm{pop}} (through, e.g., population-informed priors Fishbach et al. 2020 ; Moore and Gerosa 2021 ). In this paper, we examine the implications of this line of thought for some GW events that are currently quoted as “exceptional,” providing a quantitative analysis that complements Ref. Mandel 2026 . Exceptional events and measurement errors — Let us denote by \theta a parameter of interest of a GW event (mass, spin, …). This parameter can be understood as a draw from a detected astrophysical population \theta\sim p(\theta|\mathrm{det},\{d\}) , where \{d\} denotes data from multiple events, and \mathrm{det} refers to the fact that those events passed a detectability threshold. For each event, all we know about the parameters \theta is encapsulated in the posterior distribution p(\theta|d,\mathcal{I}) from which one often extracts summary statistics \hat{\theta} (such as mean, median, maximum a posteriori, etc.). Let us define \Delta\theta=\hat{\theta}-\theta . In general, the summary statistic will be shifted with respect to the true value, i.e. \Delta\theta\neq 0 Given a collection of j=1,\dots,N_{\mathrm{obs}} events (i.e. a catalog), an exceptional event, marked with \star , is defined as that which extremizes the summary statistic, i.e. \hat{\theta}^{\star}=\max_{j}\,\hat{\theta}^{j}\qquad{\rm or}\qquad\hat{\theta}^{\star}=\min_{j}\,\hat{\theta}^{j}\,. (1) The crucial realization is that, depending on the typical spread of p(\theta|\mathrm{det},\{d\}) and measurement errors, the “exceptionality” of an event may not only stem from the population, but also from the measurement errors. To illustrate this, suppose a single-parameter population following an exponential distribution p(\theta|\mathrm{det},\{d\})=\begin{cases}e^{-\theta}&{\rm if}\quad\theta\geq 0\,,\\ 0&\mathrm{otherwise}\,,\\ \end{cases} (2) and measurement errors that are Gaussian with a standard deviation \sigma . The population has unit expected value and unit standard deviation. The condition \sigma>1 ( \sigma<1 ) corresponds to the case where the spread induced by measurement errors is larger (smaller) than the typical spread of the population. We generate 1000 realizations of a catalog with N_{\mathrm{obs}}=100 and collect the highest mean \hat{\theta}^{\star} . Results from such a distribution of exceptional events are summarized in Fig. 1 . The main result is that, for exceptional events, the summary statistics tends to significantly overestimate the true parameter whenever the spread of measurement errors becomes compatible with that of the population. For example, for \sigma=1 , we find that about 50% of the measured parameters overshoot the true value by more than 1\,\sigma . For \sigma=3 , more than 90% of the measured values are more than 1\,\sigma away, and more than 60% are 2\,\sigma away. The latter corresponds to more than a factor of 2 of overestimation. Figure 1: Distribution of exceptional events in a catalog of N_{\mathrm{obs}}=100 using the toy population of Eq. ( 2 ) and Gaussian measurement errors. The parameter \sigma controls the importance of measurement errors compared to the width of the population. The top panel compares the measured parameter \hat{\theta}^{\star} to the true value \theta^{\star} ; the black dashed line corresponds to \Delta\theta^{\star}=0 . The bottom panel shows the distribution of measurement errors. GW231123 — We now apply this argument to the current set of GW events. We simulate catalogs by sampling from the default binary BH population of GWTC-4.0 Abac et al. 2025c and imposing a detectability threshold corresponding to a network signal-to-noise (SNR) ratio of 8. We verified our results are not affected by this specific choice. The SNR is computed using the power-spectral density and response of the detectors in the event’s observing run, which is sampled proportionally to the duration of the O1–O4 observing runs completed so far. As a cross-check, we show in Fig. 2 the events with largest total mass M_{\rm tot}^{\star}=\max_{j}M_{\rm tot}^{j} . For a catalog size comparable to GWTC-4.0 ( N_{\mathrm{obs}}=153 ), the resulting distribution peaks at about the total mass of GW231123. Figure 2: Exceptional events’ total mass distribution using the GWTC-4.0 population as described in the main text. Catalogs of different sizes N_{\mathrm{obs}} are shown as colored histograms. The thin grey distribution shows the detected population p(M_{\rm tot}|{\rm det},\{d\}) . The scatter point indicates the total mass of GW231123 obtained under agnostic priors \mathcal{I}_{\rm ag} from Ref. Abac et al. 2025a . To construct the distribution of measurement errors, we standardize the posterior distribution on M_{\mathrm{tot}} for all 153 binary BH events in GWTC-4.0 by substracting and dividing by their posterior mean. This procedure provides the relative error \Delta M_{\mathrm{tot}}/M_{\rm tot} . Given a simulated event M_{\mathrm{tot}}^{j} , the corresponding measurement error \Delta M^{j}_{\mathrm{tot}} is obtained by sampling a value from one of the 153 standardized posterior distributions selected at random with equal probability. The resulting distribution of exceptional events is shown in Fig. 3 , where we expressed results in units of \delta M_{\mathrm{tot}}=40\,\mathrm{M}_{\odot} —a value indicative of GW231123’s uncertainty on the total mass. One of the conclusions of Ref. Mandel 2026 is that the true mass of GW231123 could be below 170\,\mathrm{M}_{\odot} as opposed to the value of M_{\mathrm{tot}}=238^{+28}_{-49}\,\mathrm{M}_{\odot} obtained using population-agnostic priors, which corresponds to a shift of about 78\,\mathrm{M}_{\odot}=1.95\,\delta M_{\mathrm{tot}} . We find that less than 5% of our simulated catalogs with N_{\mathrm{obs}}=150 support this conclusion. While the typical spread of exceptional events is broader than that of the individual events, the relatively small size of \delta M_{\mathrm{tot}} with respect to the typical range of M_{\mathrm{tot}} suppresses the “exceptionality inflation.” Figure 3: Deviation of the maximum measured total mass \hat{M}_{\rm tot}^{\star} with respect to the true parameter M_{\rm tot}^{\star} in units of \delta M_{\rm tot}=40\mathrm{M}_{\odot} . Results related to exceptional events from catalogs of sizes N_{\rm obs} are shown with curves and markers. The thin grey curve indicates the corresponding distribution for the entire population p(M_{\rm tot}|{\det},\{d\}) (i.e. without maximizing). This analysis illustrates the putative “exceptionality” of GW231123’s total mass. GW241011 and GW241110 — Similarly, we now consider the component \chi_{1\mathrm{z}} of the primary BH spin aligned with the binary orbital angular momentum for the events GW241011 and GW241110 Abac et al. 2025b . Since dimensionless BH spin components are bounded to the (-1,1) interval, we expect measurement errors to strongly affect events with exceptional spin properties. We simulate GW catalogs as described above. In this case, however, posterior distributions are only shifted (and not re-scaled) by substracting the posterior mean, so that measurement errors correspond to (signed) absolute deviations, and measured values are clipped to the (-1,1) range after adding measurement errors. Results are shown in Fig. 4 . GW241011 corresponds to the highest observed aligned spin \hat{\chi}_{1\mathrm{z}}^{\star}=\max_{j}\hat{\chi}_{1\mathrm{z}}^{j}=0.66 with an uncertainty of about \delta\chi_{1\mathrm{z}}=0.08 (estimated from the reported 90% credibility interval Abac et al. 2025b , divided by 2). About 90% of our simulated catalogs with N_{\mathrm{obs}}\geq 50 report a deviation of more than 1\,\delta\chi_{1\mathrm{z}} with respect to the true value; 70% of said catalogs return a deviation beyond 3\,\delta\chi_{1\mathrm{z}} . About 10% of the catalogs return a deviation beyond \approx 8.25\,\delta\chi_{1\mathrm{z}} , which would imply GW241011 is non-spinning or even anti-aligned. GW241110 corresponds to the highest observed anti-aligned spin \hat{\chi}_{1\mathrm{z}}^{\star}=\min_{j}\hat{\chi}_{1\mathrm{z}}^{j}=-0.39 . The uncertainty in this case is about \delta\chi_{1\mathrm{z}}=0.35 Abac et al. 2025b . For N_{\mathrm{obs}}=150 , about 80% of the exceptional events are more than 1\,\delta\chi_{1\mathrm{z}} away from their true parameter values, and 70% of the exceptional events are compatible with BH spins that are preferentially aligned with the binary orbital angular momentum. Figure 4: Deviation of the largest aligned/anti-aligned spin components \hat{\chi}_{1\mathrm{z}}^{\star} with respect to the true parameter \chi_{1\mathrm{z}}^{\star} . In the top (bottom) panel, we maximize (minimize) \hat{\chi}_{1\mathrm{z}} , which is indicative of GW241011 (GW241110) and use units of \delta\chi_{1\mathrm{z}}=0.08 ( \delta\chi_{1\mathrm{z}}=0.35 ); see main text. Results related to exceptional events from catalogs of sizes N_{\rm obs} are shown with curves and markers. The thin grey curves indicate the corresponding distribution for the entire population p(\chi_{1\mathrm{z}}|{\det},\{d\}) (i.e. without extremizing). Grey areas correspond to cases where BH binaries with aligned spins are recovered with anti-aligned spins, and viceversa. Discussion — BHs with masses of \sim 30\,\mathrm{M}_{\odot} were considered exceptional ten years ago, following the detection of GW150914 Abbott et al. 2016 , because such values were much higher than those inferred from electromagnetic observations at the time. Today, BHs of \sim 30\,\mathrm{M}_{\odot} are considered vanilla Abac et al. 2025c . This evolution is bound to continue. As the catalog of detected events grows, the distribution of what is deemed “exceptional” inevitably shifts toward more extreme values. The same argument also applies to measurement errors: as we collect more and more data, it becomes increasingly likely to observe an extreme measurement error. This follows fundamentally from the use of population-agnostic priors to identify exceptional events, which is incompatible with the fact that events are deemed exceptional only with respect to a population Mandel 2026 ; Fishbach et al. 2020 ; Moore and Gerosa 2021 . In this work, we have presented a simplified model of the interplay between extreme events in a population and extreme measurement errors. We conclude that, for current catalog sizes N_{\mathrm{obs}}\lesssim 10^{3} , measurement errors drive apparent exceptional events only when their magnitude is comparable to the span of the population. This is the case for spins with uncertainties of \delta\chi_{1\mathrm{z}}\approx 0.35 , such as GW241110, but it is unlikely to be a concern for GW241011 or for the most massive event observed to date, GW231123. On the latter, our results are at odds with those of Ref. Mandel 2026 , which suggest that the total mass of GW231123 is likely to be significantly overestimated due to the use of inconsistent priors. While we agree with their overall argument, our quantitative analysis indicates that the typical uncertainty on the total mass renders this effect smaller than initially suggested. Specifically, we find that the 90% credible interval reported under \mathcal{I}_{\mathrm{ag}} Abac et al. 2025a is still likely to contain the true mass, despite the use of information that is inconsistent with respect to the population. On the other hand, our analysis suggests that GW241110, the most confidently anti-aligned spin measured in with GWs so far under \mathcal{I}_{\mathrm{ag}} ( \chi_{1\mathrm{z}}<0 at 97.7% credibility Abac et al. 2025b ), may actually not be so. Specifically, 70% of our simulated catalogs return “exceptional” anti-aligned events which are in fact non-spinning/aligned. This has major astrophysical implications, as spin is often quoted as a key smoking gun of BH binary formation channels. Echoing the key message of Ref. Mandel 2026 , our results suggest caution when interpreting astrophysically interesting events in a population-agnostic manner, as the resulting conclusions may not be robust. In addition to Ref. Mandel 2026 , we stress this caveat is important whenever measurement errors are comparable in size to the width of the underlying source population.