Politics of Science

By “politics of science” (Figure 5; keywords: ethics, education, open science, pluralism, influence, and relevance), we refer to the broad factors shaping archaeological research and its position within wider academic and public spheres. Traditionally, such concerns have been explored within theoretical archaeology. Yet the politics of science is also deeply intertwined with everyday research practice when funding bodies may be nonspecific or stipulate, for instance, open-science principles.

Open science emerged as the most frequently cited topic. Respondents emphasized the importance of free access not only to publications and data but also to reproducible research workflows (Marwick Reference Marwick2017; Marwick et al. Reference Marwick, d’Alpoim Guedes, Barton, Bates, Baxter, Bevan and Bollwerk2017). In some cases, the responses simply named this topic (and often exclusively) as the most important issue. These answers underscore the value of theorizing the preconditions and effects of open dissemination: Which initiatives genuinely improve scientific knowledge, and how can ethical and practical barriers—such as protecting sensitive site coordinates—be managed? Although these discussions reflect archaeology’s broader embrace of global open-data trends, they are especially salient in computational work, where the reproducibility of complex analytical pipelines greatly depends on transparent methods.

Education was another leading issue, given that the proliferation of computer-assisted approaches demands broader digital literacy. If theoretical archaeology is to guide the application of these new methods, coding skills and critical reflection on algorithmic assumptions must become part of basic archaeological training (Marwick et al. Reference Marwick, Wang, Robinson and Loiselle2020; Scherjon et al. Reference Scherjon, Romanowska and Lambers2019). This need is intensified by the recent public adoption of AI tools such as generative large language models (GLLMs), which are tools trained on massive text data to produce human-like text responses. Although it remains unclear whether GLLMs will most affect computational workflows or interpretive practices, several scholars (Cobb Reference Cobb2023; Tenzer et al. Reference Tenzer, Pistilli, Brandsen and Shenfield2024) predict a growing focus on digital literacy so that professional expertise can counter superficial analyses, outdated narratives, and potential “hallucinations.” Moreover, AI alignment issues—training algorithms toward specific research goals—could demand new ethical and theoretical frameworks, especially as archaeology grapples with large, potentially skewed datasets (e.g., Casini et al. Reference Casini, Marchetti, Andrea, Orrù and Roccetti2023).

Turning to ethics, participants highlighted resource inequalities (gender, funding, infrastructure) that can limit who benefits from computational tools. Collaboration offers one remedy: sharing not only data but also computational resources can help offset disparities, although such initiatives must integrate open-science safeguards to protect sensitive archaeological data.

Similarly, discussions of plurality (multivocality) reflected concern that quantitative methods might perpetuate Western scientific paradigms in the sense of foundational paradigms (Clarke Reference Clarke1968); respondents asked how to include alternative epistemologies and whether contradictory analytical results can coexist in a more inclusive interpretive space.

Although fewer in number, responses on influence and relevance noted that digital methods have the potential to reshape archaeological questions and interpretations, often having broader impacts than immediately obvious because of the ubiquity of digital and computer-based processes; for instance, 3D technologies may affect how artifacts and sites are presented to the public, or macro-archaeological research may catalyze dialogue with other disciplines (Perreault Reference Perreault2019). The underlying challenge is to ensure that these shifts promote deeper collaboration rather than reinforce existing biases. Indeed, the long-standing tradition of archaeology contributing “interesting problems” (e.g., seriation) to other fields suggests there is potential for truly bidirectional exchanges—but also for unintended oversimplifications.

Overall, the politics of science aspect reported in our survey highlights a dual imperative: harnessing the momentum of open science and digital advances while maintaining caution about ethical, educational, and plurality concerns. For computational archaeology, this means consciously embedding theoretical reflection at every stage—funding, data collection, analysis, and dissemination—to preserve archaeology’s nuanced view on the human past.

Epistemological Embedding

Computer-based methods shape, limit, and guide archaeological research. Survey responses emphasized that although digital tools can greatly enhance scholarly inquiry, they also necessitate explicit theoretical reflection—an aspect often overlooked.

A recurring theme was reflexivity, also in the sense of the implications of using computer-based methods. For example, Perry and Taylor (Reference Perry, Taylor, Matsumoto and Uleberg2018) advocated for integrating critical self-awareness into digital workflows, thereby encouraging the use of methods that foreground introspection and iterative critique.

Another focal point concerns paradigms, the ways in which existing theoretical frameworks are translated into and sometimes reshaped by computational analyses. These ways can be demonstrated by visibility analyses. They were designed around the importance of visual perception for both social interaction and individual understanding of the world (Gillings and Wheatley Reference Gillings, Wheatley, Gillings, Hacıgüzeller and Lock2020). It could also be argued that the availability of computers played a decisive role in the development of processual archaeology or GIS for landscape archaeology in the 1980s and 1990s (e.g., Allen et al. Reference Allen, Green and Zubrow1990; Clarke Reference Clarke1968).

Quantification is a prerequisite for computer-assisted methods. Participants questioned the extent to which quantification is feasible—or desirable—when dealing with complex cultural phenomena. Technical solutions, such as fuzzy logic that handles partial truths beyond simple binary true/false categories, semi-quantification, or data aggregation/assimilation that merges newly collected observational data with computational models to refine or adjust existing prediction all help in addressing ambiguity, yet their theoretical legitimacy still demands scrutiny.

Finally, the limits and possibilities of the epistemological potential of computational archaeology emerged as an overarching concern. One extreme insists that everything is quantifiable—viewing computers as able to tackle once-intractable problems—whereas those holding the opposite position contend that core aspects of human life lie beyond statistical description. Striking a balance between these views requires a systematic appraisal of computational methods.

Sources

Quantitative and computer-assisted methods often lack the built-in uncertainty that is tied to qualitative or traditional interpretations. Because data are frequently used verbatim, attention to data quality, character, and validity is crucial (35 responses to this aspect made it the second-most mentioned topic after methodology). To prevent overinterpretation, theoretically supported buffers must be deliberately incorporated.

The most obvious and most frequently mentioned aspect within this topic area is the data itself. Managing and representing data emerges as a central concern, parts of which we already discussed. A genuine archaeological challenge is handling time-related data, notably how to integrate absolute and relative dating: this is an area where theoretical archaeology could offer valuable insights. An issue relating to research design is how data collection predefines investigations. Much of our data are nominal, representing existential quantification (∃x). As one respondent remarked, “Data does not speak for itself, only in the context of scientific practices.” A specific archaeological data theory essentially specifies a general information theory. Translating general principles into field-specific necessities appears to be a desirable goal, as inferred from the responses.

Recent shifts in archaeological practice have prompted discussions about Big Data, partly spurred by Kristiansen (Reference Kristiansen2014) but also reflecting an ongoing transformation in archaeology. Although the term (De Mauro et al. Reference De Mauro, Greco and Grimaldi2015) may not perfectly fit the archaeological record, it can apply when large spatial datasets or varied types of data accumulate. Most survey responses simply named Big Data without specifying its pitfalls, yet one warned that broad-scale analysis may gloss over environmental and cultural variability. Here, the evaluation of massive datasets, not just their size, becomes the real challenge. After all, our research can be conducted easily on a multiscalar basis (e.g., Bevan and Conolly Reference Bevan, Conolly, Lock and Molyneaux2006; Hodos Reference Hodos2022).

Both big and not-so-big data carry biases from the way information is collected, processed, or distributed, which was the topic of six responses. Again, most answers simply mention this issue without much explanation. One response mentions both algorithmic bias and training data bias. This points to the fact that all evidence (inductive) is a basis for hypothesis generation and testing (deductive), so that any bias is always in the end driven by data. The response “bias through data and narratives” highlights this. Finally, the reference to “research bias including differential access” emphasizes that an unequal bias landscape also results from an unequal research landscape. We assume that the best way to minimize these biases is to compensate for such inequalities through sharing information, discussions, and academic exchange. Theoretical archaeology can help us scrutinize these blind spots and propose more equitable practises.

Validation refers to the inductive–deductive cycle and the use of “subsequent field/material data” to review the “results of computational methods” and the “validation of models.” In this way, the “reliability” of CA may be verified. It is useful to distinguish at least two levels of validation: (1) verifying that a given computational pattern or correlation is indeed present in a dataset, something that can often be established with a high degree of certainty through defined technical procedures and providing the means to reproduce the respective analyses, and (2) judging whether a specific interpretation or explanatory model accurately captures the social, cultural, or historical phenomena in question. The second level can run into “absolute truth” issues, because interpreting complex human behaviors rarely permits final proof. Instead, we can ask whether a model or interpretation is useful, robust, or consistent with known evidence in a given research context (Dewey Reference Dewey1941:178).

Uncertainty in models and data is perceived as another problem. Obviously, uncertainty must be considered when using statistical and computer models. But is there a need for a general theory of uncertainty in archaeology that goes beyond quantitative approaches? Is it even possible to develop such a theory?

Standardization is one way to mitigate some of the uncertainty by making decisions about classification at the time of data recording, rather than after data collection. However, doing so can risk oversimplification. The researcher’s primary task is to translate complex reality into scientifically manageable information, inherently simplifying it. The increasing volume of data from excavations and investigations tends to delay these necessary decision-making processes. As scientists, we are expected to make such—often imperfect—decisions and thereby to make data available for higher-level analyses. At the same time, multiple or coexisting standards can reflect a healthy multivocality, in which region-specific practices or different theoretical perspectives and research questions shape classification systems. Rather than forcing a single uniform standard, archaeology could benefit from developing robust interfaces (e.g., interoperable ontologies) that allow diverse standards to communicate with one another. It would be preferable to implement classifications and standardizations upfront while building in pathways for translation between alternative systems and providing the scientific community with the means to critically evaluate these decisions. Focusing on practical aspects of standardization—such as data formats and thesauri—is only secondarily a theoretical consideration. Every decision regarding standardized documentation affects how information is made available and becomes data for further analyses. Therefore, any standards we adopt should be explicit and transparent to enhance the reliability of the data (Reiter et al. Reference Reiter, Staniuk, Kolář, Bulatović, Rose, Ryabogina and Speciale2024).

Proxies (and by extension nominal indicators) were mentioned only a few times. This surprised us because, in our opinion, all the archaeological data are just proxy information. We suggest that the goal of archaeological research lies beyond counting and categorizing finds and features—and for all questions beyond that, the material remains represent nothing more than proxy data. Even though nominal indicators may seem more discrete than proxies, both serve to bridge interpretive theory and measurable evidence. Therefore, we consider a theory of proxies (and indicators) as a very important field of activity, possibly going far beyond what is commonly used and understood as a proxy in archaeological practice.

Challenges

It is true that accomplishing all the possible tasks discussed so far is a considerable challenge for both theoretical and computational archaeologists. However, there are specific areas, or possible assumptions and conditions of these procedures, that make them particularly challenging. Therefore, we group together 12 responses that address specific challenges.

One of these concerns the role of an individual within large datasets, the possibilities for subjecting individual social actions to quantitative analysis, and how to avoid treating exceptions or “outliers” as mere noise. Appropriate approaches to these specific questions do exist. Examples include agent-based models (ABMs) and simulations that focus on the actions of individuals as agents to assess their effect on the system as a whole (“emergence”). Theoretical questions here may be whether these approaches are sufficient, what other approaches need to be considered, and how they could better contribute to the understanding of human behavior.

A second challenge is causality or determinism. Although it is widely accepted that all historical processes are multifaceted, many archaeological explanations still reduce events to a single, overarching cause. Computational models exacerbate this tendency if they force complex phenomena into simplified causal chains. Balancing the need for a workable model with the inherent multiplicity of the past remains an ongoing dilemma.

A closely related issue is rationality, reflecting the frequently invoked assumption that humans act on the basis of complete information and clear incentives. In archaeology as in other social sciences, that people always act logically or economically should represent a baseline or reference model, rather than a blanket claim. In other words, rational models ask, “What if past agents were guided purely by cost-benefit thinking?” and then compare the outcomes to observed patterns. Such models can be useful starting points for exploring deviations from an idealized scenario. It is also important to distinguish between formal/logical rationality (i.e., internal consistency of decisions) and “reasonable assumptions” (e.g., treating humans as cost optimizers); sometimes economic theory uses those assumptions as an axiom. In reality, social or symbolic motivations can override purely economic decisions, leading to settlement patterns that defy cost-based predictions. Such discrepancies underscore the importance of incorporating cultural contexts into computational frameworks.

Finally, underdetermination was explicitly mentioned twice. This term is now primarily linked to temporal and spatial scales, where the processes that should be recorded and reconstructed must match the resolution that we can achieve with our sources. After the original publication (Perreault Reference Perreault2019) calling for a focus on large-scale questions and the analysis of large-scale data, it is certainly more of an encouragement to pursue CA more intensively. But is such a—perhaps naïve—statistical evaluation of timespans encountered in archaeological practice theoretically viable? A discussion of the topic of underdetermination can also include examining the extent to which it exists at all and can be determined.

Emerging Fields

The last overarching theme comprises emerging areas stemming from computational approaches and quantitative inquiry, along with a few “other” topics mentioned only once. We argue that such topics can potentially represent the major mainstream topics of the future.

Statistical inference was mentioned most frequently. Some respondents called for reducing the emphasis on visualization and incorporating more rigorous inferential statistics; another named the interpretation of p-values. The latter is currently being discussed in applied statistics in general (most recently, see Courtenay Reference Courtenay2024); from the archaeological point of view, it would be interesting to evaluate p-values with regard to the negative evidence in the archaeological record. Approaches that go beyond traditional p-values, such as Bayes factors and simulation-based significance tests, offer potentially richer insights. Bayes factors, for instance, enable the direct comparison of competing models, helping archaeologists weigh different interpretive frameworks (e.g., a cultural vs. environmental explanatory model) in light of the data’s likelihood under each scenario. Likewise, simulation-based methods, widely used in spatial or point-pattern analysis, can incorporate domain-specific factors like taphonomic processes, biases in site detection, or uneven sampling intensities. By generating many synthetic datasets under varying assumptions, these tests compare observed patterns to distributions that account for archaeological realities. Such tools move beyond binary “significance” to illuminate how well each model aligns with the underlying archaeological record—and they can better capture the uncertainty inherent in complex, fragmentary data.

From our perspective, models were mentioned surprisingly infrequently, with only three references. The tasks identified with regard to theory were conceptualization and validation, as well as the fact that (independent) models are suitable for establishing the link between archaeological patterns and causal processes. Providing that link is precisely where we see the strength of models: their ability to mediate the close relationship between theory and data, between deduction and induction, and between high- and low-level hypotheses. Previously, the use of models was much more widespread, particularly within processual archaeology, but archaeological theory moved away from them. This shift most likely occurred because many epistemological tools are not perceived as models at all, even though they are precisely that (Clarke Reference Clarke1972; Nakoinz and Hinz Reference Nakoinz, Hinz, Thalheim and Nissen2015). Therefore, models provide an excellent starting point for a deeper theoretical investigation of computer-assisted and quantitative archaeological practice.

Machine learning using algorithms that identify patterns in data received the same number of mentions as models in our survey. Many research and heritage practices—in archaeology, for example, the location of potential sites using remote sensing tools (e.g., Argyrou and Agapiou Reference Argyrou and Agapiou2022) or the classification of archaeological objects (e.g., Korokhina Reference Korokhina2024)—are currently being revolutionized. The AI boom will also create a surge in advanced machine-learning approaches. Before using those approaches, it is necessary to develop a theorization at an early stage before it is surpassed by reality. The supposed objectivity of machine-learning methods needs to be questioned from a theoretical point of view (Tenzer et al. Reference Tenzer, Pistilli, Brandsen and Shenfield2024), as does the bias in the archaeological data itself. The same can be said about automation.

The final topic that was mentioned more than once is complexity, often framed as an antidote to deterministic narratives. Complexity research recognizes emergent phenomena as facts and is thus well suited to archaeological research. Another facet is the complexity of archaeologically observable past societies and of the conditions and effects of their emergence (e.g., Daems Reference Daems2021). However, its core assumptions also demand careful theoretical and methodological grounding within archaeology.

Finally, a range of other topics—including aDNA, navigation theory, Bayesian statistics, and multimaterial additive manufacturing—reveal the extensive domain in which computational archaeology now operates. Although many originate from other disciplines, archaeology’s theoretical frameworks must adapt to—or critically evaluate—such specialized innovations. The widespread sense that some questions remain “theoretically underdeveloped” underscores a collective desire for more explicit reflection on the interplay between novel methodologies and long-standing archaeological aims.

Summary

Questions concerning methodology, sources, and data were considered to be the most important. In absolute terms, various aspects of middle-range theory, suitability of methods, data, integration, and research design stood out. In the area of the politics of science, open science was the topic most mentioned. Next were issues about bias, Big Data, the individual, and education. All other aspects mentioned were represented in the third order with a maximum of four mentions. Although the research efforts of many computational archaeologists are theoretically very well informed, theoretical archaeology can significantly contribute to many areas of computer-assisted and quantitative archaeology.

What More Has Computational Archaeology to Offer?

CA is well integrated into archaeological research and, from a certain perspective, currently seems to dominate the field: most archaeologists use some kind of computer applications, and many investigations are not feasible without the use of computer-aided techniques ranging from documentation, both in the field and labs, to data analysis and visualization. We are also convinced that CA can grow in its responsibility and abilities to handle the current “grand challenges” of archaeology (Kintigh et al. Reference Kintigh, Altschul, Beaudry, Drennan, Kinzig, Kohler and Limp2014) while not losing sight of the need to reflect on the epistemological embedding of its methods, applications, and the politics of science.

One of the advantages of CA is its potential to bridge the gap between quantitative and qualitative research approaches and scales: critical and reflective approaches based on sociocultural anthropological standpoints should be integrated into CA, given its general openness toward any kind of datasets and its wide range of methodological possibilities (Perry and Taylor Reference Perry, Taylor, Matsumoto and Uleberg2018). For example, anthropological understandings of kinship can be integrated into collections and analyses of aDNA datasets (e.g., Brück Reference Brück2021), and reflective thinking can be incorporated into the premises used for ABM (e.g., Wunderlich and Laabs Reference Wunderlich, Laabs, Heitz, Wunderlich, Hinz and Furholt2023). The potential to connect different levels of scale that we observe in CA is thus linked to the potential to connect different disciplines.

Another important aspect is that archaeology ultimately finds evidence in patterns of material culture distributions. For example, CA is frequently used in spatial analysis to find and highlight statistical patterns and distributions that provide information on past spatial use (e.g., Gillings et al. Reference Gillings, Hacigüzeller and Lock2020). Computational methods may extend the interpretation of data beyond assessing the validity of results and the potential “strength of signal” suggesting patterns of spatial use: one could also use that data to recognize and understand a change in human behavior and to determine to what extent present-day archaeological views on change may align with those of past peoples (Crelin Reference Crelin2020).

Conversely, there are multiple ways in which theories and theoretical approaches can be integrated into quantitative approaches of CA. We agree with Perry and Taylor (Reference Perry, Taylor, Matsumoto and Uleberg2018) that theoretical approaches that focus on the reflections of one’s self, one’s data, method, or the ideology of a given scientific paradigm should play a far stronger role in CA than is currently the case. After all, quantitative approaches need to be backed up and reflected on by qualitative data and theoretical approaches. Hypothesis-generated research questions must lead to an investigation seeking to answer it, which means engaging in the deductive–inductive cycle. Computer-based quantitative studies should be question driven and thus theory driven.

Open Science

Open science (OS), as already mentioned, is becoming essential and in archaeology must also include the theoretical considerations that lead to interpretations and narratives. However, it is strongly linked to the digitalization and computerization of science in general (Science Europe and German Research Foundation 2018), and its principles (Open Science 2023)—open methodology, open source, open data, open access, open peer review, and open educational resources—do not necessarily cover the need to open up the theoretical part of science. This may be sufficient for many natural sciences that are based on the “scientific method” per se and do not follow any specific or only generally accepted schools of thought. They may also have less need to think about the epistemological foundations of their scientific methods, because humans as acting subjects are less central to their work.

This is not true for humanities nor for archaeology. To make sense of the material remains we excavate and to turn them into knowledge, we need a whole chain of data transformations before we can infer human action. This chain of inferences must necessarily use deductively constructed top-down assumptions that we must use as gap fillers. Of course, the openness about theory can be redundant and need not be formulated anew in each publication, but the basic models that play a major role in one’s own research agenda should be stated. In our effort to make sense of the past, we deal with uncertainty not only in data analysis but also in those basic models we employ. Therefore, we need to make our middle-range theories explicit.

Of course, CA is strongly involved in pushing novel agendas into archaeology, which requires analytical procedures and datasets to be explicit and available. CA has the possibility to be a role model in archaeological science and open many existing black boxes. It should also be a role model for collaboration, transparency of methodology, data sharing and accessibility, as well as good and reproducible scientific practices (see also Perry and Taylor Reference Perry, Taylor, Matsumoto and Uleberg2018). Because CA studies might be under suspicion of skewing data and using methods to produce results that fit initial research questions and narratives, many of them already follow the ideas of OS and publish their datasets and methodology following the rules for reproducible research. This is demonstrated, for example, by the ever-growing list of papers using and publishing R code in Ben Marwick’s “CRAN Task View: Archaeological Science” (Marwick et al. Reference Marwick, Wang, Giusti, Crema, Galili, Peare Bartholdy and Spake2022). Archaeology would benefit from adopting most OS principles, because the preparation and processing of most archaeological data are of central importance, whether advanced computational methods are used or a written description is the result. A task for a theoretically oriented CA (or a CA-oriented theory) would therefore be to devise ways to reveal precisely these theoretical foundations in the same manner as is evolving today with the methodology of analyses.

In terms of practical implementation, a range of open-science guidelines, platforms, and tools already exist to help archaeologists make their work more accessible and reproducible. For instance, the Open Science Framework (OSF) offers free, versioned hosting for data, code, and project documentation. “The Turing Way” provides a broad-based handbook on reproducible research in data science, which is largely adaptable to archaeological workflows. Open Context and the Journal of Archaeological Data offer curated, peer-reviewed data sharing specifically for archaeology, and Marwick et alia’s (Reference Marwick, d’Alpoim Guedes, Barton, Bates, Baxter, Bevan and Bollwerk2017) discussion of OS practices in archaeology and the “CRAN Task View: Archaeological Science” (Marwick et al. Reference Marwick, Wang, Giusti, Crema, Galili, Peare Bartholdy and Spake2022) demonstrate how reproducible code and data can be integrated into research design.

Caution Is Needed

What became clear in the survey results is that CA is perceived as being a field that tends toward positivism and believes in the superiority of ever larger datasets and smarter algorithms. There is some truth to this view. However, we must also remember that often the data itself and the theoretical model associated with it inspire the notion that any data analysis must lead to significant results and that the more sophisticated the method used for data analysis, the more credible it is. It therefore seems necessary to promote processes for evaluating and validating the data and theories, as well as the methodology used. This applies to both quantitative and qualitative approaches, as well as to research projects conducted on a large or small scale. Bias cannot be eliminated in any kind of archaeological research, but it is equally present in quantitative and qualitative approaches.

Developing a Common Language

Something that should unite all archaeologists is the desire to create a representative image of past societies based on our most up-to-date knowledge, regardless of the spatial or temporal extent of the past under investigation. This may be more of an ideal than a reality, but it should be an ideal that most of us agree with. To achieve it, we use all kinds of archaeological data and transform this information into a narrative using different methods and theories.

Archaeology is a discipline that engages in many ways with issues of multidisciplinary and public interest. As such it needs to seek a common language with experts from different disciplines, as well as the public. Researchers and other members of academia who are actively involved in continuous education can approach a common language through education. Common vocabularies can be developed and shared through the educational systems offered by universities, and the extensive possibilities offered by summer schools, online tutorials, and online courses provide a basis for developing increased understanding between researchers from different disciplines. Despite the opportunities that arise from shared understanding, consideration must be given to the obstacles that stand in the way of sustainable implementation.

The teaching of archaeology at universities covers in most cases a wide range of necessary topics and competences, and it would be unrealistic to expect every single institute and department to offer an extensive range of courses specifically on CA and archaeological theory. Yet, basic knowledge in both areas should be firmly anchored in the undergraduate curricula. As Garstki (Reference Garstki2022) points out, a study program should challenge digital and other archaeological skill sets regularly so students can achieve literacy in them. A very concrete but serious obstacle to digital literacy is the lack of coding skills among most archaeologists. In other disciplines, such as astronomy or biology, an introduction to basic programming techniques and data science is now a standard part of undergraduate education. Why should a science as open and interdisciplinary as archaeology lag behind? In the age of AI, basic digital literacy will grow in importance in working with future CA approaches. As mentioned earlier, it will take an effort on all sides to bridge the gap that currently seems to divide at least parts of the archaeological community.

From Reflection to Action

The discussions outlined here highlight deep interconnections between theoretical archaeology and computational practice. In this section, we translate these connections into more concrete, step-by-step suggestions. Although some themes—like open data or transparent workflows—are relevant to many scientific disciplines, we highlight here how they can be adapted to the context and challenges of archaeological research.

A major step forward is to integrate theoretical reflection directly into research design, ensuring that computational methods are neither applied in isolation nor chosen purely for their novelty. Rather than formulating research questions only after collecting or modeling data, researchers benefit from co-developing hypotheses with both theoretical specialists and computational experts at the very outset of a project. Early collaboration helps prevent situations where methods are retrofitted to existing datasets; it simultaneously clarifies the epistemological frameworks in which data and models will be interpreted. For instance, when planning a GIS-based project on regional settlement patterns, inviting experts on social agency and landscape phenomenology to initial workshops can refine the choice of parameters (e.g., anisotropic cost surfaces or agent-based simulation of daily movement) and ensure that relevant theoretical assumptions such as the role of past perceptions of space are carried into the computational design.

Agent-based models are a prime example of incorporating theory early in CA; one example is Chliaoutakis and Chalkiadakis’s (Reference Chliaoutakis and Chalkiadakis2020) agent-based trading model for Bronze Age Crete. In their approach, autonomous, utility-seeking household agents occupy a virtual landscape modeled on the Minoan world around Knossos, enabling researchers to investigate how socioenvironmental factors shape intersettlement networks. Crucially, the authors tested two spatial interaction submodels (XTENT and Gravity) to see how theoretical assumptions about settlement “importance” (e.g., prestige, population, or longevity) affect trading ties. They found that privileging importance over proximity (via the Gravity model) aligned more closely with archaeological evidence of Minoan site hierarchies and the disruptions sparked by the Theran volcanic eruption. By examining network properties like centrality and clustering, the study highlights how theory-driven computational designs can reveal fresh insights into the organization, adaptability, and resilience of past societies.

Meaningful collaboration, however, depends on an adequate level of digital literacy across the archaeological community. To build this capacity, university curricula can include structured instruction in basic programming (e.g., R or Python), as well as hands-on introductions to open data repositories or agent-based modeling platforms. Practitioners who are already working in the field can benefit from short, targeted workshops, potentially offered by digital humanities centers or professional associations, that pair conceptual debates about rational versus nonrational decision-making with practical coding exercises. For example, a workshop might first address how cultural beliefs influence Bronze Age trade patterns and then guide participants through coding an agent-based model in NetLogo that incorporates nonrational decision rules. By coupling theoretical discussions with direct experience in implementing models, such activities encourage archaeologists to see computational methods less as black boxes and more as flexible tools subject to critical scrutiny.

Central to these critical and reflective approaches is the commitment to transparent, open, and collaborative practices. Archaeological data are inherently shaped by layers of interpretive decisions—ranging from field documentation conventions to classification systems—and making datasets and code available for reuse and replication can help demystify these processes. Researchers may publish software scripts, annotated data tables, or 3D models on platforms such as GitHub, Zenodo, or the Open Science Framework, accompanied by clear project documentation and references to theoretical assumptions. In cases where data sensitivity is a concern (e.g., in revealing precise site locations), strategies like sharing coarsely gridded coordinates or requiring formal data-use agreements can balance reproducibility with ethical obligations. When made open, these computational workflows illuminate the choices involved in data cleaning, classification, or model parameterization—allowing both specialists and nonspecialists to evaluate the validity and potential biases of the resulting analyses.

Although this article cannot provide an exhaustive manual for implementing open science, it is important to emphasize that practical, detailed guidelines already exist. In addition, archaeologists can tap into several established frameworks to simplify and standardize their adoption of open-science practices. For instance, the FAIR Data Principles (Findable, Accessible, Interoperable, Reusable) offer clear criteria for structuring and archiving datasets, and the UNESCO Recommendations on Open Science provide a global roadmap for ethically sharing research outcomes. In Europe, the ARIADNE infrastructure illustrates effective data curation and repository models. Aligning local initiatives with these or similar frameworks can accelerate the integration of open-science principles, fostering a more cohesive and collaborative archaeological community.

To further reduce the risk of hidden or unintended biases, it is helpful to institutionalize reflexivity within the entire research workflow from project design to publication. One practical approach is to include short, explicit “reflexivity statements” that identify the theoretical stance, the known limitations of datasets, and potential uncertainties or interpretive leaps. These statements can also outline how the results will be “ground-truthed” against secondary datasets, ethnographic analogies, or historical records. For example, a predictive site-location model might be validated by random checks in the field, discussions with community stakeholders, or the close reading of textual evidence. Such iterative cross-checking goes beyond merely reporting error margins or p-values; it acknowledges that interpretive frameworks shape how we collect and evaluate data in the first place.

Finally, theoretical pluralism can be nurtured by consistently trying out multiple interpretive or computational scenarios to ensure that no single approach—processual, postprocessual, evolutionary, or otherwise—dominates the narrative unchallenged. CA, in particular, offers opportunities for comparing how different theoretical assumptions translate into diverging model outcomes. By systematically experimenting with alternative interpretations, such as contrasting a rational-actor view of resource use with a culturally embedded perspective, archaeologists can reveal both the implicit norms that govern a model’s construction and the patterns that persist across various assumptions. This openness to multiple working hypotheses is especially valuable for illuminating the complexities of human behavior hidden behind aggregate patterns in large-scale datasets.

Taken as a whole, these interconnected strategies—integrating theory early, improving digital literacy, adopting transparent practices, fostering reflexivity, and championing pluralism—represent an evolving framework for promoting innovation in CA. They go beyond merely acknowledging the importance of theory or data sharing, instead insisting that practices like cocreated research design or reflexivity statements become routine. Although not every project can implement all these measures, even incremental progress—such as publishing a modest reflexivity note or sharing code with a clear README—can shift the culture toward a more critically engaged, ethically grounded, and interdisciplinary mode of inquiry.