2. Origin explanations according to Ebersole and Shrewsbury

Reflecting on origin of life studies, Ebersole and Shrewsbury (Reference Ebersole and Shrewsbury1959) examine what it means to formulate origin explanations. Questions about origins, they argue, come in two main types. Contrast for instance (A) the question of the origin of a particular social custom, such as the custom men used to follow to walk on the outside of streets when accompanied by women, with (B) the question of the origin of social customs in general. The questions will receive different explanations: An explanation of (A) may run as “In the Middle Ages, refuse was thrown from house windows. Someone walking close to the building was less likely to be splattered with it than someone walking on the outside. Hence, men wishing to protect the dignity and finery of their women, took to walking always on the outside. It became an item of chivalry, and men who did not observe the courtesy were criticized,” while an explanation of (B) will be that “At a certain time, people do things for what is, or is believed to be a good reason. These things become sanctioned, and violators are criticized, ostracised, or even killed. Because of the conservative nature of man in his society, these modes of behaviour are preserved even after the reasons non longer apply and are never again given, after no sanctions remain except the reminder, ‘That’s just the way it’s done’” (Reference Ebersole and Shrewsbury1959, 105).

Ebersole and Shrewsbury’s diagnostic is that origin explanations of type (A) work by appealing to “facts of history,” while origin explanations of type (B) involve a “general plan or pattern” (Reference Ebersole and Shrewsbury1959, 105–6). Though each of these answers to origin questions appears to be a perfectly satisfactory type of origin explanation, the authors note a close relationship between the two: The general plan or pattern of type (B) explanations is tacitly assumed to be at work in explanations of type (A). In other words, the historical facts mentioned by (A) explanations follow the general patterns of (B) explanations.

Ebersole and Shrewsbury argue that, unlike social customs which have clear cases of (A) and (B) origin explanations, explanations of the origin of life seem to waver between the two explanatory modes. On the one hand, the question typically refers to a unique historical event—the emergence of life on Earth around four billion years ago—and it receives tentative answers of a similarly historical nature, such as reconstructions of early environmental conditions. On the other hand, it is also approached through abstract, pattern-based models drawn from physics and chemistry, for instance, by invoking general chemical pathways that could lead to the formation of complex organic molecules. This duality introduces a further complication, as Ebersole and Shrewsbury point out: There remains a conceptual gap between what is being explained—the formation of certain organic molecules—and what we truly want to understand—the origin of life.

Although the distinction they draw between these two explanatory types (A) and (B) is valuable (and I will return to it in section 5), I contend that Ebersole and Shrewsbury’s account ultimately misses the mark. It fails to capture what is distinctive about origin explanations as such. Their framework suggests that origin explanations either consist in assembling historical facts or in applying general explanatory schemes. While the first type resembles early forms of historical or narrative explanation, and the second aligns with the deductive-nomological (DN) or inductive-statistical (IS) models, neither, I argue, captures what is specific about explanations that are genuinely about origins. More sharply put, Ebersole and Shrewsbury show that origin explanations can be subsumed under two broad types of explanatory models that apply to a vast array of phenomena—but in doing so, they overlook what makes origin explanations specific. If such explanations are no more than standard historical or covering-law explanations, then what justifies their persistent appeal across domains? What legitimizes the use of origin terminology in the first place? The beginning of an answer, I suggest, lies in recognizing that an origin explanation targets a particular kind of explanandum—what I will call an “origin phenomenon.” In what follows, I will clarify what counts as such a phenomenon. Then, I will argue that explanations of origin phenomena exhibit a distinctive epistemic feature: They work by identifying a bottleneck in the causal network that leads to such origin phenomena.

3. Origin phenomena

The exploration of origin questions encompasses a wide array of phenomena that I propose to refer to as “origin phenomena.” It goes without saying that not all phenomena are proper targets for origin questions. Many physical, chemical, biological, and social phenomena—while certainly legitimate objects of explanation simpliciter—do not intuitively invite inquiries into their origins. Consider, for example, physical effects such as rainbows, electromagnetic laws like Ampère’s law, superconductivity, or general relativity’s account of gravitational distortion. Certainly, these phenomena can all be targets for explanations but framing them within the context of an origin explanation does not seem like the right type of approach. The same holds for many chemical reactions, biological processes, or sociological dynamics that are typically explained by uncovering the mechanisms, laws, or models that govern their operation. The nature of these phenomena is such that one would not intuitively think of them as origin phenomena in need of origin explanations.

A possible diagnostic criterion for these cases lies in their ahistorical nature. These phenomena are typically investigated and explained through repeated experiments or observations. In this respect, their occurrence at a specific moment in the history of the world is irrelevant to the explanation-seeking question we entertain about them. Instead, the quest for explanation concerns the intelligibility of the very unfolding of the phenomena, as opposed to their origin. In contrast, origin phenomena are marked by a form of historical situatedness, typically with a discernible start—the origin. Consider the example of social customs, such as the custom of walking on the outer side of the pavement that Ebersole and Shrewsbury mentioned: One can certainly inquire about this practice in terms of origin, presupposing a time before the practice existed. The same holds for phenomena like the emergence of the French Resistance during World War II, the outbreak and spread of the COVID-19 pandemic, or the appearance of life on Earth. These are paradigmatic cases in which origin questions seem not only natural but pressing: All exhibit some form of historical situatedness that seem inherent in origin phenomena.

Yet not all historical events qualify as origin phenomena. Many, in fact, are ill-suited to be framed in this way. For instance, Descartes’s 1633 decision not to publish The World in light of Galileo’s condemnation is historically grounded and explainable (Gaukroger Reference Gaukroger1995)—but asking for the “origin” of that decision feels off-target. One might explain the decision narratively, but the event appears too isolated to count as an origin phenomenon: It lacks the scale and structural development we typically associate with such phenomena. For a phenomenon to qualify as an origin phenomenon, I will argue, it must be of sufficient scale and complexity—typically involving a causal expansion over time.

The civil movement of the French Resistance exemplifies this: It arose historically in the early 1940s from the French civil society, gained traction following de Gaulle’s Appeal of 18 June 1940, and gradually evolved into a more-or-less coordinated network of intelligence-gathering, sabotage, clandestine communication, armed resistance, and civil disobedience (Gildea Reference Gildea2015; Wieviorka Reference Wieviorka2016; Jackson Reference Jackson2018). The Resistance expanded across time, participants, actions, and strategic influence. A similar point holds for the COVID-19 pandemic. It likely began with a limited zoonotic spillover event but escalated rapidly into a global health crisis, involving international spread, government policies, economic disruption, and changes in social behavior (Ye et al. Reference Ye, Yuan, Yuen, Fung, Chan and Jin2020; Banerjee et al. Reference Banerjee, Doxey, Mossman and Irving2021; Lytras et al. Reference Lytras, Xia, Hughes, Jiang and Robertson2021). Life on Earth too exemplifies this characteristic: Most scientific accounts begin with a lifeless early Earth, followed by a progression of chemical evolution during which chemical compounds complexified, protocells emerged, and a gradual diversification and expansion of living forms developed, ultimately giving rise to the vast integrated biosphere we observe today (Smith and Morowitz Reference Smith and Morowitz2016; Sutherland Reference Sutherland2017; Szostak Reference Szostak2017; Nghe Reference Nghe2025).

In all three cases, the phenomena can be seen as occupying relatively distinct and expanding causal regions of the world, characterized by internal causal density and external distinctiveness—what makes individuation of the phenomena possible in the first place. A significant characteristic of what we are drawn to call origin phenomena, I propose, is that they can be represented as causal networks capturing regions that not only emerge and stabilize but also expand in scope and significance over time. One way to visualize this is as a temporally unfolding directed graph: Nodes represent causal relata, and arrows capture the causal links connecting them across successive moments. In the case of an origin phenomenon, such a causal graph would be expected to display a marked increase in both the number and connectivity of nodes as time progresses.

Formally, such structures can be cast as directed acyclic graphs (DAGs), of the sort reconstructed in causal inference frameworks (Pearl Reference Pearl2000; Spirtes et al. Reference Spirtes, Glymour and Scheines2000) and widely employed in interventionism and structural equation approaches to causation (Hitchcock Reference Hitchcock2001; Woodward Reference Woodward2003). These frameworks offer precise tools for representing causal dependencies, especially when causal relata are treated as variables connected through structural equations. Yet in many domains—including the historical sciences—causation can be studied without commitment to such formal machinery. In these cases, causal relata may be conceived not as values of variables but as events (Kim Reference Kim1973; Lewis Reference Lewis1986), facts (Mellor Reference Mellor1995), or states of affairs (Armstrong Reference Armstrong1997), while causation may be understood along alternative perspectives as afforded by diverse counterfactual theories (Lewis 1973; Reference Lewis2000), causal process theories (Salmon Reference Salmon1998), or conserved quantity approaches (Dowe Reference Dowe2000; Kistler Reference Kistler2006), not to mention more intuitive and practice-based conceptions.

For instance, in contexts where narrative explanation is central, events often provide the most fitting causal relata, particularly when paired with counterfactual analyses. Beatty’s account of evolutionary possibility exemplifies this approach, relying on counterfactual events and their dependencies (Beatty Reference Beatty2017). What matters for my argument is that, once actualized and traced across temporal stages, these counterfactual dependencies yield causal networks. Such networks, when represented as directed graphs, make visible the temporal unfolding that characterizes origin phenomena.

Consider also cases like pandemics or the diffusion of social habits, where the central task is to trace how phenomena propagate through individuals. Such processes are often modeled as contact networks, with nodes representing individuals and edges capturing pathways of transmission or influence, thereby revealing communities and hubs that structure the spread (e.g., Cauchemez et al. Reference Cauchemez, Bhattarai and Marchbanks2011; Kumar et al. Reference Kumar, Oke and Nahmias-Biran2021). Yet they can equally be represented as temporally unfolding directed graphs, where each time-slice indicates who infects or influences whom. This dynamic representation makes their expansion over time explicit and illustrates a distinctive feature of what we are drawn to call origin phenomena.

For these reasons, by “causal networks” I do not mean to restrict the analysis to DAGs in the narrow sense of causal inference theory. Rather, I use the term more broadly to designate structured sets of causal dependencies, however construed. In this broader sense, causal networks serve as the scaffolding through which origin phenomena can be identified across domains, from molecular pathways to social dynamics. What matters for the account advanced here is that such networks can be represented as or translated into temporally directed graphs because it is in this form that the topological features of origin phenomena become most salient.

Of course, the relevance and scale of what counts as significant within such networks varies depending on the epistemic context, including the framing of the inquiry, the granularity of the research question, and the level of explanation sought. This context-sensitivity explains why de Gaulle’s Appeal of 18 June, while often understood as one of many triggering events for the Resistance, does not by itself constitute an origin phenomenon at the same high-level scale, especially when compared to the Resistance. The Appeal was a singular event, not a sprawling causal process. Yet, in a finely grained epistemic context—for example, one concerned with political communication or the psychology of leadership—the Appeal might be shown to fit within a reasonably well-delineated and expanding causal network at that very fine scale. In that context, the speech becomes the phenomenon of interest, and its origin may turn out to be a meaningful question. In this way, origin questions, like all explanation-seeking questions, are shaped by the epistemic context in which they are asked. Whether something qualifies as an origin phenomenon depends not just on its historical emergence but also on its perceived significance, expansion, and internal causal density relative to the aims and granularity of inquiry.Footnote ² In other words, an origin phenomenon is not merely something that began and grew, but something that did so in a way that matters considering a specific epistemic context.

Thus, I propose to construe an origin phenomenon in the following way:

This view acknowledges that, given the structuring role played by the epistemic aims of inquiry, the identifiability of origin phenomena rests on their historical situatedness and their causal development.

4. Origin explanations and causal bottlenecks

I propose to think of an origin explanation as a specific kind of explanation: One that targets an origin phenomenon as just defined. Given that origin phenomena are typically understood in causal terms, origin explanations are, in turn, typically causal explanations: They situate the causal network underpinning the origin phenomenon within a broader upstream causal network. What distinguishes them, I argue, is that they end up revealing a bottleneck in the causal network: a narrow convergence region from which a proliferating causal trajectory unfolds—that downstream network corresponding to the origin phenomenon. This bottleneck, I contend, is a defining—and possibly unique—feature of origin explanations. In other words, an origin explanation explains by showing how an origin phenomenon results from a convergence of causal lines and initiates a downstream expansion of causal interactions (see Figure 1).

Figure 1.

Causal bottleneck in an origin explanation. An origin phenomenon (blue/squares) is characterized by a lasting and expanding causal network. An origin explanation (orange/dots) identifies a preceding causal region that converges at a bottleneck—interpreted as the origin of the phenomenon—before proliferating downstream. Additional parts of the causal network are suggested in gray/white dots.

To explain an origin phenomenon is thus to identify the portion of the broader causal network that converges toward the causal structure underpinning the phenomenon, thereby revealing a bottleneck. As with the characterization of origin phenomena, the epistemic context here too determines the appropriate level of causal detail (i.e., its epistemic granularity), including how far back one must go to produce an explanatorily adequate account. I thus propose the following characterization of an origin explanation:

Under this framework, it is crucial to note that an origin explanation isn’t just any causal explanation; rather, it is a causal explanation tailored to elucidate an origin phenomenon as per the characterization outlined in the preceding text (section 3). In doing so, an origin explanation uncovers a bottleneck within the causal landscape. The origin of an origin phenomenon is precisely that bottleneck:

A natural question is what, exactly, characterizes the causal links that constitute a bottleneck. Importantly, causes at bottlenecks are not to be understood in terms of sufficiency, that is, as a small set of factors jointly guaranteeing the production of the phenomenon. Nor should they be reduced to an arbitrary or pragmatic choice of salient causes. Rather, bottlenecks are defined structurally, as convergence points within a temporally unfolding causal network. Such a network is identified at a particular level of granularity, one that is appropriate to the research question and the way the explanandum has been framed as an origin phenomenon. Within that context, all causes in the network already carry explanatory salience; what distinguishes those at the bottleneck is their structural position. Bottlenecks are sites where otherwise diverse or proliferating causal lines are funneled through a narrower set of dependencies, making subsequent developments hinge upon them. The causes located at such junctures thereby acquire a particular epistemic status: Their significance derives not from intrinsic strength or necessity in isolation, but from their role as connectors within the network’s topology. This confers on them a special interventionist relevance. They may serve as strategic levers for preventing undesirable origin phenomena—such as in public health interventions designed to block zoonotic spillovers—or for intentionally triggering new ones, as in laboratory research aimed at reconstituting minimal pathways to life. Thus, bottlenecks mark privileged sites where explanation, prediction, and intervention can be brought into alignment.

To illustrate this, consider the (very) simplified historical reconstruction of the emergence and development of the French Resistance during World War II. As previously argued, in a suitable epistemic context (e.g., World War II historiography), the Resistance qualifies as an origin phenomenon—it did not exist prior to a certain time and unfolded into a significant and expanding causal network. Explaining its origin involves tracing an upstream causal network composed of wartime events, the unique role of Charles de Gaulle, the broadcast of the Appeal of 18 June, and other historically salient factors. This upstream causal network intersects with the downstream network of the Resistance, and that intersection reveals a bottleneck—a key moment or set of conditions that initiated the emergence of the Resistance (Figure 2A). In this sense, explaining the origin of the Resistance is a paradigmatic case of origin explanation.

Figure 2.

Bottlenecks in origin explanations. (A) Origin of the French Resistance during World War II. The causal network of the Resistance (blue/squares) connects backward to multiple causal threads: de Gaulle’s trajectory (yellow/lozenges), the 18 June Appeal (green/hexagons), and broader wartime events (orange/dots). (B) Origin of life. Three causal strands converge—a vesicle formation pathway, the emergence of a replicase ribozyme, and molecular linkers—followed by diversification of living entities.

A similar structure appears in origin-of-life research. Though far from settled, leading hypotheses propose that life emerged from the convergence of three prebiotic systems: self-assembling vesicles, self-replicating RNAs, and molecular linkers—likely other RNAs—enabling integration and coupling (Sutherland Reference Sutherland2017; Joyce and Szostak Reference Joyce and Szostak2018; Benner et al. Reference Benner, Bell and Biondi2020; Nghe Reference Nghe2025). These systems emerged from preexisting prebiotic chemistries (Figure 2B).Footnote ³ The explanation of life’s origin is thus framed as a convergence of causal lines resulting in the bottleneck of protocell emergence, followed by an expansion of living entities. Again, the explanation qualifies as an origin explanation because it locates and accounts for a causal bottleneck initiating a distinctive and expanding phenomenon—life.Footnote ⁴

This pattern, I propose, generalizes across a wide range of origin explanations in a variety of domains.Footnote ⁵ What characterizes them is not merely their historical situatedness and their causal character, but how they work: To explain an origin phenomenon is to show how this origin phenomenon arose through a bottleneck in the broader causal network of the world.

5. From tokens to types: Generalizing origin explanations

The bottleneck framework developed so far applies most directly to token origin phenomena—that is, specific, individual instances of origin events. Yet, as rightfully noted by Ebersole and Shrewsbury (Reference Ebersole and Shrewsbury1959), origin phenomena can also occur in multiple instances of the same type. Consider again the example of social customs: One might ask about (A) the origin of a specific custom, such as walking on the outer side of the pavement, or (B) the origin of customs as a general category. This distinction reveals two possible foci for origin questions: either on particular origin events in their specific contexts, or on general patterns shared across multiple instances of similar origin phenomena.

Take the case of zoonotic diseases—infectious diseases transmitted from animals to humans. Many specific instances of zoonoses, such as outbreaks of the rabies virus, have been well documented. Each of these constitutes a proper token origin phenomenon: a specific event occurring at a particular time and place, embedded in a concrete causal context. Each could, in principle, be the subject of a distinct origin explanation, as outlined in the preceding text. However, these cases often exhibit recurring causal structures. Rabies infections, for instance, commonly follow a pattern involving a reservoir host, pathogen spillover, human exposure, and disease transmission (Rupprecht et al. Reference Rupprecht, Hanlon and Hemachudha2002; Hampson et al. Reference Hampson, Dushoff, Bingham, Gideon Brückner and Dobson2007). These recurrent features suggest that the origin of rabies infections can be explained not only individually but also at a more general level—by reference to the causal pattern that characterizes their emergence as a type.

Furthermore, researchers have identified similarities across different zoonoses caused by distinct pathogens. For example, the causal pathways of rabies and Ebola outbreaks both involve analogous upstream factors such as host species distribution, pathogen prevalence, and human-animal interaction (Plowright et al. Reference Plowright, Parrish and McCallum2017). Such cross-pathogen similarities suggest the possibility of an even more abstract level of origin explanation—one that targets the type-phenomenon of zoonotic disease emergence in general (Figure 3). These regularities can be used to identify intervention measures to constrain the flow of pathogens, typically at bottleneck regions. They have also informed generalized models and frameworks for predicting and explaining zoonotic outbreaks (Ruan et al. Reference Ruan, Wen, He and Wu2021), as well as high-level pandemic emergence models. Crucially, even at this generalized level, what grounds the explanatory relevance of such models is their ability to identify bottlenecks—convergent points in causal space, such as pathogen release events—that recur across different instances of the type. These bottlenecks highlight the distinctive feature of origin explanations.

Figure 3.

Causal bottlenecks in the origin of zoonoses. Influence of major causal factors (left axis) on the emergence of human infectious diseases (bottom axis) from animal sources (top axis). The variables on the left (e.g., “reservoir distribution,” “pathogen prevalence”) are coarse-grained aggregates of finer causal processes. Each oval represents a stage through which causal flow must pass, with its size indicating both the probability of passage and the effective number of causal connections. The diagram thus combines a high-level causal sequence with a schematic of how finer causal networks contract and expand around bottlenecks (adapted from Plowright et al. Reference Plowright, Parrish and McCallum2017, with permission).

Following Ebersole and Shrewsbury, then, it makes sense to acknowledge that origin explanations can also target what one may call “type origin phenomena,” and can do so by appealing to patterns that emerge from analyzing multiple related token origin phenomena. In such cases, the explanation operates by generalizing over similarities in their upstream causal networks. Nevertheless, the feature that remains distinctive of origin explanations—at both token and type levels—is their focus on revealing bottlenecks in causal networks. This motivates characterizing type origin phenomena as follows:

To explain a type origin phenomenon is to detect and generalize over the upstream causal similarities across its instances. These include recurring causal factors and relations that, together, define a convergent causal structure. I will thus define a type origin explanation as follows:

What sets type origin explanations apart from generic pattern-based or statistical explanations is that they remain anchored in the logic of origin explanations: They aim to reveal recurrent bottlenecks that give rise to expanding downstream causal networks.

To illustrate this further, consider again the question of the origin of life. One might ask: (A) How did life arise on Earth?—a question that targets a specific origin phenomenon. But one might also ask: (B) How does life emerge in general?—a question that targets a type origin phenomenon, should we one day identify life on Mars, on Enceladus, or on exoplanets outside our solar system. Although the existence of such extraterrestrial life remains hypothetical, ongoing research in astrobiology and exoplanet biosignature detection (Schwieterman et al. Reference Schwieterman, Kiang and Parenteau2018; Enya et al. Reference Enya, Yoshimura, Kobayashi and Yamagishi2022; Wong et al. Reference Wong, Bartlett, Chen and Tierney2022; Dannenmann et al. Reference Dannenmann, Klenner and Bönigk2023; Malaterre et al. Reference Malaterre, ten Kate and Baqué2023; McMahon and Cockell Reference McMahon and Cockell2024; Seager et al. Reference Seager, Welbanks, Ellerbroek, Bains and Petkowski2025) prepares the groundwork for answering such a question. If multiple instances of life’s origin are found, a general pattern might emerge involving, for example, the synthesis of prebiotic building blocks, the emergence of informational molecules (e.g., replicating RNA), vesicle formation, coupling between information and compartmentalization, and metabolic complexification. The aim of fields such as universal biology is precisely to identify general principles or “laws of life” that could hold across diverse instances of life, independent of their specific biochemical makeup (Sterelny Reference Sterelny1997; Mariscal and Fleming Reference Mariscal and Fleming2018; Cleland Reference Cleland2019). In such a scenario, the question of life’s origin could be formulated not just at the level of token origin phenomenon (A) but also at the level of a type origin phenomenon (B), open to explanation by identifying shared bottlenecks across different instances. These might include key transitions such as the onset of Darwinian evolution or the emergence of autocatalytic networks (Higgs and Lehman Reference Higgs and Lehman2015; Davies and Walker Reference Davies and Imari Walker2016; Goldenfeld et al. Reference Goldenfeld, Biancalani and Jafarpour2017; Adamski et al. Reference Adamski, Eleveld and Sood2020).

As with token origin explanations, type origin explanations are grounded in the targeting of origin phenomena—whether individual tokens or broader types. Token origin explanations trace the specific upstream causal network leading to a single origin event. In contrast, type origin explanations uncover causal patterns by identifying similarities across the upstream networks of multiple token origin phenomena. In both forms of explanation, the identification of causal bottlenecks remains central, serving as the key to understanding how origin phenomena emerge.

6. Three possible objections

7. More on the bottleneck account

Before concluding, let me to elaborate on two interrelated facets of the bottleneck account: the epistemic status of the causal networks invoked, and the structural variability of bottlenecks themselves.

A characteristic feature of many origin phenomena across disciplines is that they lie in the past and are often only accessible through indirect evidence. As a result, origin explanations frequently rely on traces—for example, fossil, chemical, genetic, or behavioral—left by the phenomena in question. However, these traces are often scarce, degraded, or ambiguous, making the reconstruction of causal networks leading up to the phenomena highly partial or even underdetermined (Tucker Reference Tucker1998; Stanford Reference Stanford2001; Turner Reference Turner2005; Currie Reference Currie2021; Malaterre Reference Malaterre2024). This raises a question about the status of the causal networks invoked in the bottleneck account: Are these networks to be understood as actual historically instantiated structures, or as possible configurations compatible with available evidence?

My view is that the causal networks referenced in the bottleneck account should be construed as idealized representations, shaped by the epistemic constraints and investigative resources of a given context. They represent what one would aim to reconstruct if one had access to an optimal evidential base. In this ideal case (Figure 4A), the upstream portion of the causal network would include all pertinent antecedent conditions, processes, and interactions, while the downstream portion would detail the ramifications and outcomes of the phenomenon. In practice, however, the evidential landscape is rarely, if ever, ideal. Data are often sparse, unevenly distributed, or noisy. The connections among available causal factors may be ambiguous or subject to competing interpretations. Accordingly, the reconstructed causal network will often be highly fragmentary or “sketchy” (Figure 4B). Yet even under such conditions, scientists routinely engage in the task of inferring plausible networks, typically by combining empirical constraints with reasoned hypotheses. This then often results in outlining the broad contours of likely causal networks (Figure 4C).

Figure 4.

Epistemic status of the causal network in origin explanations. (A) Ideal causal network given a specific epistemic context. (B) Sketchy causal network inferred from available evidence. (C) The same sketchy causal network supplemented by hypothesized supplementary causal factors and relationships, revealing a bottleneck in causal space.

For example, although bats are strongly suspected to be natural reservoirs of the Ebola virus, the specific mechanisms of viral maintenance and transmission within bat populations remain poorly understood (Plowright et al. Reference Plowright, Parrish and McCallum2017; Riesle-Sbarbaro et al. Reference Riesle-Sbarbaro, Wibbelt and Düx2024). In many documented Ebola outbreaks, the upstream causal factors—ecological conditions, host behavior, viral shedding—remain elusive, as do the precise circumstances of initial human exposure. This results in a fragmentary network of causal claims and possibilities (hence the presence of question marks in Figure 3). Still, enough evidence exists to posit that a set of converging causal factors—reservoir species, ecological triggers, human encroachment—likely contributed to a transmission bottleneck that enabled viral spillover. The existence of such a bottleneck—however loosely defined—is inferred not from a complete causal map but from a partially reconstructed network in which certain nodes and relationships—despite the gaps—appear pivotal.

The case of the origin of life is also illustrative. No satisfactory explanation yet exists. Even the downstream causal structure—what followed once life emerged—is incompletely understood. While molecular phylogenetics points to a Last Universal Common Ancestor (LUCA), LUCA appears to be far more complex than the first truly living systems, implying that large segments of the early biosphere remain undiscovered or lost (Cornish-Bowden and Cárdenas Reference Cornish-Bowden and Luz Cárdenas2017). Upstream, evidence is even thinner. Geological and planetary data sketch a highly uncertain picture of early Earth conditions around 4 billion years ago, while microfossil evidence places the existence of life at least by 3.4 billion years ago, possibly earlier. Experiments in prebiotic chemistry have shown how certain molecular building blocks might form under early Earth-like conditions, and analyses of meteorites have revealed the presence of extraterrestrial organics (Cleaves et al. Reference Cleaves, Chalmers, Lazcano, Miller and Bada2008; Sutherland Reference Sutherland2017; Oba et al. Reference Oba, Takano and Furukawa2022). But the pathways from this chemical diversity to something recognizably alive are still missing. Moreover, early prebiotic systems may have left no traces due to subsequent geological processes. In all likelihood, the origin of life will not be discovered in the traditional sense but will have to be reinvented (Eschenmoser Reference Eschenmoser2007). For all these reasons, any explanation of the origin of life will likely consist of how-plausible—or even how-possible—scenarios blending historical constraints, prebiotic experiments and engineered proof-of-concepts syntheses. Even so, many such scenarios converge on a constricted set of transitional conditions—bottlenecks—marking the emergence of life from nonlife.

A second key facet of the bottleneck account concerns the form that bottlenecks take or should take to qualify as proper bottlenecks as per the proposed framework. I do not mean the framework to require that origin explanations should all resemble a neatly delineated, hourglass-shaped structure, with upstream factors funneled through a narrow transitional region before symmetrically diverging downstream. In practice, bottlenecks exhibit considerable variability in shape, scope, and structure.

For one thing, the causal networks identified as relevant to origin phenomena do not need to be strictly delineated and separated from the rest of the world’s causal structure. Rather, the goal is to extract subsets of causal space in which factors are more densely interconnected with each other than with external variables. This is a qualitative rather than strictly formal criterion: What matters is the relative clustering and salience of causal relationships, not the strict causal isolation as formalized for instance by Markov blankets in probabilistic causal models (Pearl Reference Pearl2009). I propose to adopt a looser approach: identifying a causally salient region embedded within a larger system, given the explanatory aims of a particular epistemic context.

Similarly, the “shape” of the bottleneck is likewise qualitative. Bottlenecks need not be narrow points or symmetric arrangements: They can be regions or constriction patterns with diverse topologies. Still, scientists can often identify a region of transition or compression—a zone through which many causal chains must pass, and whose structure determines the shape of what follows. Just as in real life, not all bottlenecks look alike, but we generally recognize one when we see it.

That said, these qualitative intuitions are not incompatible with quantitative formalization. Network science provides several tools to analyze and detect bottlenecks in complex systems. For instance, the Cheeger constant (Mohar Reference Mohar1989) can measure the extent to which a network contains a sparsely connected partition—an indicator of a potential bottleneck. Other indicators such as betweenness centrality allow one to quantify the structural importance of nodes through which many connecting paths pass (Newman Reference Newman2018). Thus, the bottleneck account remains compatible with both qualitative intuition and quantitative network analysis.