4.1. Types of legal evidence

In legal contexts, direct evidence directly speaks to what is at stake in a criminal case. When evaluating whether a defendant is guilty of committing a crime, examples of direct evidence are the defendant’s confession or eyewitness testimony of someone directly observing the crime. By contrast, circumstantial evidence speaks to what is at stake via an intermediate inferential step. Lagnado (Reference Lagnado, Dawid, Twining and Vasilaki2011) and Fenton et al. (Reference Fenton, Neil and Lagnado2013) illustrate the evidentiary contributions of direct and circumstantial evidence through Bayesian networks capturing the structure of legal arguments. Consider the network in figure 1 as an example.

Figure 1.

Network for the criminal investigation. G: defendant is guilty. O: defendant was at the scene. W1: testimony first witness. W2: testimony second witness.

The investigation evaluates whether the defendant is guilty of robbing a shop (G). Testimonies of two witnesses (W1, W2) have been obtained. W1 testifies that they saw the defendant threatening violence and obtaining money from the cash register. W2 testifies that they saw the defendant at the shop prior to the crime occurring but then left the place themselves. In the network, each node represents a variable that can take on different values.Footnote ¹ The directed edges (arrows) between the nodes represent dependencies between variables. For example, the directed edge between G and W1 represents that W1 probabilistically depends on G. The testimony W1 is more likely to obtain if G is the case. According to Fenton et al. (Reference Fenton, Neil and Lagnado2013, 90), in such a network, “direct evidence involves a single causal link from the issue to be proved to the evidence. If true, this evidence proves the hypothesis in question.”

By contrast, Fenton et al. (Reference Fenton, Neil and Lagnado2013, 91) point out that “circumstantial evidence … is linked to the issue to be proved via a causal path involving at least two steps, for instance, as evidence for motive or opportunity.” According to the network in figure 1, W2 provides circumstantial evidence for G. That is because W2 testifies to the fact that the defendant had the opportunity to rob the shop. After all, they were at the scene. This is represented by variable O. In turn, O obtaining is directly probabilistically relevant to G: The defendant being at the shop makes it more likely (though not certain) that they committed the robbery.

Figure 2.

Network for the stromatolite case studies. L: ancient microorganism activity. S: contemporary stromatolite-like structures. M: ancient marine environment. C: contemporary chemical signature.

Neither G nor O is knowable as such; they are hypotheses that must be evaluated in light of the available evidence (Fenton et al. Reference Fenton, Neil and Lagnado2013, 18). While G is the ultimate hypothesis under investigation, O takes on the status of an auxiliary hypothesis that needs to be established for the circumstantial evidence to become relevant to G. This illustrates the additional inferential step needed in cases when legal evidence is circumstantial rather than direct.

4.2. Types of traces

What I propose to call direct traces are the remnants of the past that the causal, informational, and evidentiary understandings of traces are concerned with. Because these views characterize only one type of trace, they do not come with “directness” built in explicitly. In fact, it even seems as if many of the epistemic situations of historical science resist using the term direct in relation to traces. Trace-based reasoning, or so the argument might go, seems anything but direct. However, direct in the present context should not be understood in a physical sense, that is, as two things (events, objects) being close together in space and time. Traces may be spatially close to long-past events, but they are certainly not temporally close.

Rather, in analogy to the legal distinction, direct traces are direct in two senses. The resulting view of direct traces is close to the characterization on the evidentiary view (Currie Reference Currie2018). First, there is a direct causal chain from the historical event in question to the trace. This is less restrictive than Fenton et al.’s (Reference Fenton, Neil and Lagnado2013, 90) condition that there be “a single causal link” to account for the longer length and ramifications of the causal chain when dealing with events in the deep past. The second sense of “directness” concerns the inference from trace to (a feature of) the historical research target. Direct traces directly probabilistically depend on a hypothesis about the historical research target being the case. This is what makes the inference from one to the other a “direct” one. Note, however, that this does not mean that this inference is easy or infallible. As it is most prominently specified in the evidentiary view of traces, such inference requires specific and sometimes hard-to-establish pieces of background theory to be in place.

Let’s look at instances of direct traces in the study of (putative) stromatolites. In both the cases from Australia and Greenland, researchers are concerned with whether the activity of ancient cyanobacteria—our “suspects” so to say—formed discernible sedimentary layers in a certain location. The most important direct traces in these cases were the morphological indicators in the rock record: the peculiar shapes and forms that motivated giving a closer look to the rock formation’s origins. If these are traces of life, they are direct traces. There is a direct causal chain from the past event, the activity of ancient microbes, to a present-day phenomenon, the occurrence of layered and peculiarly shaped columns of rock. Moreover, with the right background knowledge in place, the existence of stromatolite-like structures directly depends on the past activity of microbes. Relevant knowledge in the stromatolite cases concerns, for example, knowledge about contemporary stromatolite-forming microbes.

To explain what I understand by circumstantial traces, let me give a simplified network reconstruction of the factors involved in the stromatolite cases (figure 2).

In this network, the connection between variables S and L corresponds to finding a direct trace of ancient microorganism activity. Variable C represents a piece of evidence that constitutes a circumstantial trace relative to L. Relative to M, however, C is a direct trace. C probabilistically depends on M; M is probabilistically relevant to L: It is more likely that ancient microorganisms were active in a location if (features of) the environmental setting allowed them to thrive, such as in the shallow marine environment evidenced by the particular contemporary geochemistry. Note that in this network, S and C represent empirical evidence, while L and M represent hypotheses about the past. In the case studies, the ultimate hypothesis to be evaluated, the focus of the investigation, concerns L. Then, M is needed to make possible the extra inferential step needed to connect C, the circumstantial trace, to L, the research target in the deep past.

The point just made concerns the epistemic contribution of circumstantial traces. Considering the previous discussion, this might give rise to a couple of additional questions about the nature of circumstantial traces. First, what kind of evidence qualifies as a circumstantial trace? Second, what justifies calling circumstantial traces traces in the first place? Answering these questions requires adding some explicitly causal considerations to the foregoing epistemic considerations. Circumstantial traces are traces of (collections of) events that preceded a particular historical target event and causally contributed to it. In the framework that I am proposing, the connection between past events and circumstantial traces is not a direct causal chain. However, the two are not unrelated; they are part of the same (subset of a) causal network by sharing an earlier common cause. In the network in figure 2, contemporary stromatolite-like structures (S) and contemporary chemical signatures (C) share a common cause in the past. In summary, then, a circumstantial trace must fulfill two conditions:

Present-day phenomenon A is a circumstantial trace of past event B iff …

1. i. A and B share a common cause in the past (causal condition) and

2. ii. A allows an inference to B via an intermediate inferential step (epistemic condition).

Both conditions are needed. The causal condition on its own would let in too much. Every contemporary remnant of a past event shares an earlier common cause with some other remnant somewhere in the deep past. That does not mean, however, that one can actually use these remnants in a historical reconstruction. The solution is the additional epistemic condition. But the epistemic condition on its own would not allow us to distinguish between traces and other types of evidence. The connection in the network for the criminal case also corresponds to causal relations, and with the circumstantial evidence implicating something like motive or opportunity, my causal condition is met. However, requiring that “circumstantial evidence … is linked to the issue to be proved via a causal path involving at least two steps” (Fenton et al. Reference Fenton, Neil and Lagnado2013, 91) allows for different causal links than the one I specified, for example, evidence of postoffense conduct of a defendant.

Circumstantial traces can function both in a positive way, giving further support for a hypothesis about the past, and in a negative way, putting into question such a hypothesis. The former applies to the investigation of the Strelley Pool Chert stromatolites. These sedimentary structures are now widely held by researchers to be the result of early life activity. While investigating the corresponding rock structures, researchers mapped out the geological setting in which the structures were found. They closely analyzed various features of what Allwood et al. (Reference Allwood, Walter, Kamber, Marshall and Burch2006) refer to as the paleoenvironment, among them the contemporary chemical signatures found at the site which I explicitly built into the preceding network (C). These were one of the focal points of the investigation by van Kranendonk et al. (Reference van Kranendonk, Webb and Kamber2003) and also played a role in gaining insights into the past geochemical setting in Allwood and colleagues’ (Reference Allwood, Walter, Kamber, Marshall and Burch2006, Reference Allwood, Walter, Burch and Kamber2007) research. These contemporary signatures are direct traces of an ancient marine environmental setting. In the way they are used in the reconstruction of early forms of life, they are circumstantial traces of ancient microorganism activity. In this case, a hypothesis about the environmental setting (M) serves as an auxiliary hypothesis that allows for connecting ancient microorganism activity to its circumstantial traces.

Note that connecting contemporary chemical signatures (C) to an ancient marine environment (M) is not a trivial task. It depends on pieces of relevant background knowledge that are, for instance, also highlighted by Currie (Reference Currie2018) in the evidentiary understanding of traces. However, connecting features of the ancient environmental setting (M) to the hypothesis about ancient microorganisms (L) is not trivial either. In their investigation of the Pilbara rock structures, Allwood et al. Reference Allwood, Walter, Kamber, Marshall and Burch(2006) considered the environmental setting that contemporary living stromatolites thrive in a nearby locality to foster this link. Hence there is considerable work involved in establishing both whether M could be the case and whether there is a relationship between M and L. In the Pilbara stromatolite case, both these connections were positively established. Because of this, the contributions of S and C point in the same direction: They both indicate the activity of ancient microbes on the Earth in the past. Let’s say that if the contributions of direct and circumstantial traces come together in this way, they are in alignment. If direct and circumstantial traces align, they are both positively relevant for the historical research target in question, even though only one of them is directly causally downstream from the research target. Alignment in this sense is not a passive feature of the evidence but must be actively established and argued for in the course of an investigation in the way just highlighted.

In the Isua Greenstone Belt investigation, it is much less clear whether the direct and circumstantial traces are in alignment in the way just described. The team of researchers supporting the stromatolite interpretation cite, among other things, geological evidence in the surrounding area that is indicative of an ancient marine environment (Nutman et al. Reference Nutman, Bennett, Friend, van Kranendonk and Chivas2016). Let’s unpack this in light of the distinction between direct and circumstantial traces. The further evidence invoked by both Nutman et al. and their critics can be categorized as circumstantial traces. However, in this case, considering the circumstantial traces did not, at least not unequivocally, support the hypothesis put forward by Nutman et al. This is because, to the extent that circumstantial traces in the Greenland case do support particular features of an environmental setting, those are ultimately not positively relevant to establishing a hypothesis about the presence of ancient microbes in the locality. To invoke the comparison to the legal case again, imagine an investigation in which establishing that the defendant had opportunity fails. This could be, for instance, because a witness is giving the defendant an alibi for the time of the crime indicating that the defendant could not actually have been present at the crime scene. In a related way, stromatolite-forming microorganisms are less likely to have been active in the past if they did not have the opportunity either, that is, if a suitable environmental setting was not in place for them to evolve and thrive, as Allwood et al. (Reference Allwood, Rosing, Flannery, Hurowitz and Heirwegh2018) and Zawaski et al. (Reference Zawaski, Kelly, Orlandini, Nichols, Allwood and Mojzsis2020) argue. Note, however, that the stromatolite case is much less clear than providing the alibi in the crime case might be. Again, determining whether direct and circumstantial traces align is not a trivial task, and it will not be self-evident from the given traces.