Archives and science have long been treated as separate domains: the archive as a retrospective storehouse of traces, and science as a forward-facing enterprise of discovery, verification and innovation. Yet this distinction is increasingly untenable. Scientific knowledge depends on archives not simply because archives preserve evidence, but because they organise the conditions under which evidence can be recognised, trusted, circulated, reanalysed and transformed into public knowledge. The archive is therefore not a passive container but an epistemic infrastructure: it decides what may count as data, how data are described, who may access them, what forms of interpretation become legitimate, and which silences or exclusions remain embedded in the record. From colonial repositories to digital data portals, from open citation databases to algorithmic audit trails, archives shape the production of scientific truth by mediating between memory, authority and evidence. To understand archives scientifically, and science archivally, is therefore to recognise that knowledge does not emerge from isolated facts but from institutional systems of classification, preservation, metadata, access and reuse.
Achille Mbembe’s account of the archive provides a crucial starting point because it refuses to reduce archives to collections of documents. For Mbembe, the archive is inseparable from the building, the institution, the rituals of access and the authority that decides which fragments of life deserve preservation. The archive turns selected documents into proof, granting them a status they did not possess before; it transforms ordinary papers into public evidence through classification, secrecy, custody and eventual consultation . This has profound implications for science, since scientific evidence also depends on institutional legitimation. A dataset, specimen, field note or experimental record does not become authoritative merely by existing. It must be stabilised through protocols, metadata, repositories, peer review, calibration and citation. In this sense, scientific archives participate in what Mbembe calls an instituting imaginary: they create the impression of continuity, totality and collective inheritance, even though every archive is necessarily partial. The scientific archive, like the state archive, gathers debris from the past and arranges it into a form that can speak in the present.
Ann Laura Stoler extends this insight by arguing that colonial archives should be read not only as sources but as subjects of inquiry. Her phrase “along the archival grain” is especially important for thinking about archives and science. Rather than extracting information from colonial records as though they were transparent windows onto the past, Stoler asks scholars to examine the genres, repetitions, classifications, silences and bureaucratic habits through which colonial states produced knowledge . This method is directly relevant to scientific archives, because scientific records also contain epistemic conventions. Laboratory notebooks, biodiversity databases, climate models, archaeological reports and genomic repositories do not merely report reality; they encode assumptions about measurement, relevance, classification and uncertainty. To read a scientific archive “along the grain” is to ask how its categories were made, why certain observations were repeated, which standards became authoritative, and which forms of knowledge were excluded because they did not fit institutional formats. Such a method complicates the ideal of scientific neutrality by showing that evidence is always organised through historically situated infrastructures.
The problem of classification is central here. Colonial archives transformed populations into administrable categories; scientific archives often transform phenomena into measurable, comparable and reusable data. This transformation is necessary, but never innocent. Stoler shows that colonial commissions on “poor whites” in the Dutch East Indies did not simply describe social problems; they produced racial, moral and administrative categories through the very act of documentation . Similarly, contemporary scientific datasets produce their own categories of visibility. A biodiversity database may privilege species that are easier to observe, regions with better funding, or taxonomies inherited from colonial natural history. A medical dataset may encode assumptions about race, sex, disability or normality. A climate archive may reflect the uneven geography of sensors and satellites. The archive does not only preserve what science knows; it helps define what science is able to know. This is why archival critique is not external to scientific method. It is part of scientific accountability.
Digital archives intensify these questions because they promise openness while multiplying new forms of mediation. Borgman, Scharnhorst and Golshan’s study of DANS, the Dutch Data Archiving and Networked Services institute, shows that digital data archives are not passive repositories but knowledge infrastructures: robust systems of people, artefacts and institutions that enable data sharing and reuse . Their research demonstrates that open data does not become usable simply because it is uploaded. Contributors must prepare files, create metadata, decide access conditions and negotiate credit; consumers must search, interpret and sometimes request permission; archivists must curate, migrate, document and preserve datasets. Figure 2 in their article maps this mediation among data contributors, repositories, user interfaces, search engines, metadata harvesters and research communities, making visible the network of actors behind apparently simple data access . In scientific terms, this means that reproducibility depends not only on data availability but on infrastructural labour. Without metadata, preservation standards and human assistance, “open” data may remain technically accessible but epistemically unusable.
The quality of metadata is therefore a scientific issue, not merely an administrative one. Nogueras-Iso, Lacasta, Ureña-Cámara and Ariza-López argue that Open Data portals often prioritise rapid publication while neglecting the quality of the descriptions that make datasets discoverable and reusable . Their work adapts an ISO 19157-based method to evaluate Open Data metadata structured through the W3C DCAT vocabulary, assessing dimensions such as completeness, logical consistency, temporal quality, thematic accuracy, positional correctness and free-text clarity. This framework shows that bad metadata is not a minor defect. If titles are vague, formats inconsistent, dates incoherent, spatial references inaccurate or access URLs broken, the scientific value of the data collapses. A dataset without reliable metadata is like a specimen without provenance: it may exist, but its evidentiary force is weakened. Page 10 of their article is particularly instructive because its workflow diagram shows how automated quality checks produce quantitative and conformance results, while its temporal figure demonstrates how publication, modification and validity dates must align for datasets to remain trustworthy . Metadata quality is thus a condition of scientific reliability.
This infrastructural view of archives also reframes the politics of open science. OpenCitations, as described by Peroni, offers an important example of how scholarly communication itself can become an open archival system. Citation data are often treated as secondary traces of research, but they structure discovery, reputation, bibliometrics and research evaluation. Peroni argues that open citation data must be structured, separable, identifiable, available and legally reusable, so that citation relations can be audited, queried and reused outside proprietary databases . In scientific practice, this matters because citations are not neutral ornaments. They create maps of intellectual debt, influence and legitimacy. A closed citation index centralises control over scholarly memory, while an open citation infrastructure makes the relations among publications more transparent and reproducible. Open citations therefore function as an archive of scientific association, revealing how knowledge travels, clusters and gains authority. They also demonstrate that the politics of archives extend beyond historical documents into the metrics and infrastructures that govern contemporary science.
However, archives can also reproduce exclusion under the language of participation. Ulises Ali Mejias’s critique of digital networks is useful here because it shows that inclusion in digital systems can deepen inequality when participation is captured by capitalist infrastructures . His concept of nodocentrism describes a world in which only what appears as a node in a network becomes visible, while what remains outside the network is rendered illegible. Scientific archives face a similar risk. Data that are digitised, indexed and made interoperable become visible to search engines, funding bodies and researchers; data outside these systems may disappear from scientific attention. This is particularly significant for indigenous knowledge, local environmental observations, non-English scholarship, informal archives and community science. The digital archive can expand access, but it can also create new hierarchies between data that are machine-readable and data that are socially meaningful but technically marginalised. Open science must therefore ask not only whether data are online, but whose data, in whose formats, governed by whose standards, and for whose benefit.
Dourish and Bell deepen this critique by showing that infrastructure is never merely technical. Their analysis of ubiquitous computing argues that space itself is organised through infrastructures of practice, culture, memory and power . They distinguish between the experience of infrastructure, when systems become visible through use or breakdown, and the infrastructure of experience, through which everyday life becomes meaningful. Scientific archives operate in precisely this double mode. They usually recede into the background when functioning well, but they shape the experience of knowledge by determining what researchers can find, compare, trust and reuse. When a repository fails, a link breaks, metadata are missing or access is denied, the archive suddenly becomes visible. But even when invisible, it structures scientific possibility. Archives are therefore not afterthoughts to research; they are part of the spatial, technical and cultural environment in which science happens.
John Durham Peters’s philosophy of elemental media widens this point by suggesting that media are not just devices or channels but environments that make communication possible . His image of the cloud is especially productive for thinking about scientific archives. The cloud is not an immaterial space of pure storage; it is a metaphor that conceals servers, energy systems, legal jurisdictions, interfaces and corporate ownership. Scientific data stored “in the cloud” may appear universally available, but they remain dependent on material infrastructures and institutional arrangements. Peters’s elemental approach reminds us that archives are atmospheric as well as architectural: they surround knowledge practices, enabling them so routinely that their mediation can be forgotten. Scientific archives are part of the weather of research. They create the conditions under which knowledge circulates, condenses and precipitates into evidence.
Algorithmic systems introduce another archival challenge: the preservation and interpretation of error. Mike Ananny argues that algorithmic mistakes should be understood as sociotechnical events rather than isolated technical glitches . His case study of remote proctoring shows how a facial detection system with higher error rates for darker-skinned students revealed not just a flawed algorithm but broader racial, socioeconomic and pedagogical assumptions. For archives and science, this matters because algorithmic systems increasingly classify, retrieve, rank and interpret scientific data. If their errors are not archived, audited and made public, they disappear into private troubleshooting. Ananny’s call to “see like an algorithmic error” suggests that scientific archives should preserve not only successful outputs but also failures, anomalies, uncertainty, model revisions and disputed classifications. Errors can become public problems and scientific resources when they reveal structural assumptions. The archive of science should therefore include the traces of breakdown, not merely the polished products of success.
Taken together, these perspectives show that archives and science are bound by a shared problem: how to make knowledge durable without making it falsely complete. Archives stabilise evidence, but stabilisation can become exclusion. Science requires standardisation, but standards can erase situated meanings. Open data promises reuse, but reuse depends on labour, metadata and context. Digital repositories promise access, but access is mediated by platforms, policies and power. The archive is therefore both enabling and dangerous: it preserves traces against oblivion, but it also disciplines them into authorised forms. Scientific responsibility lies in making this mediation visible. Researchers must cite data, document methods, preserve uncertainty and respect the communities from which knowledge emerges. Archivists must balance openness with protection, automation with craft, and preservation with interpretability. Institutions must fund not only data production but also long-term stewardship. Without such commitments, archives become symbolic gestures rather than functioning infrastructures of knowledge.
In conclusion, archives are not external to science; they are among science’s most important conditions of possibility. They preserve evidence, organise memory, mediate access, stabilise trust and make reuse possible across time. Yet they also select, classify, exclude and govern. A critical theory of archives and science must therefore move beyond the idea of the archive as storage and recognise it as an active epistemic system. Mbembe teaches us that archives confer status on debris; Stoler shows that archives produce governance; Borgman and colleagues reveal the mediating labour of digital data archives; Nogueras-Iso and colleagues demonstrate that metadata quality determines openness; Peroni shows the importance of open citation infrastructures; Mejias warns that networks can create invisibility through inclusion; Dourish and Bell show that infrastructures shape experience; Peters reminds us that media are elemental environments; and Ananny shows that errors can become public knowledge. Science depends on all these archival operations. The future of trustworthy knowledge will therefore depend not only on better instruments or faster computation, but on more just, transparent and sustainable archives.
Bibliography
Ananny, M. (2022) ‘Seeing Like an Algorithmic Error: What are Algorithmic Mistakes, Why Do They Matter, How Might They Be Public Problems?’, Yale Journal of Law & Technology, 24, pp. 342–364.
Borgman, C.L., Scharnhorst, A. and Golshan, M.S. (2018) ‘Digital Data Archives as Knowledge Infrastructures: Mediating Data Sharing and Reuse’, revision submitted to Journal of the Association for Information Science and Technology, 28 September.
Dourish, P. and Bell, G. (2007) ‘The infrastructure of experience and the experience of infrastructure: meaning and structure in everyday encounters with space’, Environment and Planning B: Planning and Design. Published online 9 March.
Mbembe, A. (2002) ‘The Power of the Archive and its Limits’, in Hamilton, C., Harris, V., Taylor, J., Pickover, M., Reid, G. and Saleh, R. (eds.) Refiguring the Archive. Dordrecht, Boston and London: Kluwer Academic Publishers, pp. 19–26.
Mejias, U.A. (2013) Off the Network: Disrupting the Digital World. Minneapolis and London: University of Minnesota Press.
Nogueras-Iso, J., Lacasta, J., Ureña-Cámara, M.A. and Ariza-López, F.J. (2017) ‘Quality of Metadata in Open Data Portals’, IEEE Access. doi: 10.1109/ACCESS.2017.DOI.
Peroni, S. (2022) ‘OpenCitations: a short introduction’, ULITE-ws: Understanding Literature References in Academic Full Text at JCDL 2022, CEUR Workshop Proceedings.
Peters, J.D. (2015) The Marvelous Clouds: Toward a Philosophy of Elemental Media. Chicago and London: University of Chicago Press.
Stoler, A.L. (2002) ‘Colonial Archives and the Arts of Governance’, Archival Science, 2, pp. 87–109.
