Archives make growth visible and maintenance invisible. We count publications, identifiers, datasets, records and links because accumulation is easy to measure. Yet every addition creates obligations. Definitions become outdated, links disappear, repositories change interfaces, duplicate records proliferate and superseded formulations remain searchable long after they cease to represent current thinking. A knowledge system therefore grows twice: once through what it adds and again through the maintenance work produced by those additions. The second growth is less visible, although it determines whether the first remains useful. A persistent identifier can continue resolving perfectly while pointing to an obsolete record. A document can survive while the relationships that once made it intelligible have become unclear. A copied version can remain accessible while nobody knows whether it is canonical. Preservation and maintenance are therefore different achievements. Preservation keeps an object available; maintenance keeps its condition interpretable. This requires more than technical repair. Version histories have to show how definitions change. Broken links need correction. Metadata must be harmonized as schemas evolve. Duplicates should be identified without erasing legitimate archival copies. Relations require review. Some formulations eventually need to be marked as superseded or retired. Retirement is especially important because digital systems encourage a false equivalence between persistence and permanent validity. An outdated definition may remain historically valuable, but its status should be explicit. Maintenance also imposes an ethical limit on scale. A corpus can expand faster than its maintainers can review it. At that point accumulation produces opacity rather than knowledge: more records exist, but fewer can be confidently interpreted. This is maintenance debt, whether or not the term is formalized. The crucial point is that research architectures should be designed in relation to the labour required to sustain them. Infrastructure studies have long shown that systems become most visible when they fail. Knowledge infrastructures are no exception. Broken links, contradictory versions and abandoned records reveal the work that had previously remained hidden. A more mature scholarly culture would treat that labour as part of knowledge production itself. Maintenance is not the administrative residue that follows research. It is the continuous practice through which research remains usable, attributable and historically coherent.
Bowker, G. C. and Star, S. L. (1999) Sorting Things Out. MIT Press.
Jackson, S. J. (2014) ‘Rethinking Repair’, in Media Technologies.
Star, S. L. and Ruhleder, K. (1996) ‘Steps Toward an Ecology of Infrastructure’, Information Systems Research, 7(1), 111–134.
DataCite (2021) DataCite Metadata Schema Documentation, Version 4.4.
Nogueras-Iso, J. et al. (2017) work on metadata quality in open data portals.