Tropical conditions in the Istmo-Colombian area severely affect pottery preservation, limiting diagnostic features available for conventional classification. Meanwhile, large-scale archaeological projects have produced detailed sherd-level datasets that are typically forced into pre-existing typologies or, where no prior framework exists, into new typologies shaped by the knowledge domain of each project. Conventional ceramic classification in the region has relied primarily on geographic criteria and on well-preserved specimens that are rarely representative of the preservation of the pottery assemblages, with little attention to technological attributes. This study proposes an unsupervised clustering approach that builds on these newly available datasets to overcome the classification challenges derived from severe environmental erosion and the sparse decoration of these pottery assemblages. We used a database of approximately 25,000 ceramic fragments within a chronological range of 400 BCE - 1500 CE recovered from the Mina de Cobre Panamá project to develop a quantitative data-driven classification approach based on low-visibility technological attributes. The workflow combines Gower distance computation, clusterability testing, and HDBSCAN clustering to identify technological clusters. The analysis revealed five technological clusters, each characterised by distinct temper composition signatures. Archaeological validation through spatial distribution, co-occurrence analysis, and chronological association with 45 radiocarbon dates suggest that these clusters probably correspond to geographically coherent and temporally structured communities of practice.
Sánchez-Gómez et al. (Wed,) studied this question.