Daejeon, July 27–31
Digital textual criticism now operates at two scales. At one end, major institutional projects like the Editio Critica Maior (ECM) have transformed our understanding of textual transmission through methods like the Coherence-Based Genealogical Method (CBGM) (Mink 2004; Gurry 2017). These projects depend on substantial infrastructure—dedicated servers, IT staff, and bespoke Virtual Manuscript Rooms (VMRs), such as the NTVMR, for transcription and collation (Houghton 2016).
At the other end, independent scholars assemble ad hoc pipelines: an XML editor for transcription, command-line scripts for tokenization, CollateX for alignment, and separate tools like Lera for visualization (Dekker et al. 2015). While these components exist, the workflow remains fragile. Each handoff introduces friction and potential data loss.
This paper describes Apatosaurus, a Virtual Research Environment developed during doctoral research on the Greek manuscripts of John of Damascus (Flood 2024). The tool’s architecture—built on Progressive Web App (PWA) technology—addresses specific infrastructural limitations encountered in decentralized manuscript research. Currently supporting a growing international community of scholars across various manuscript traditions, the tool demonstrates that local-first architectures can effectively bridge the gap between ad hoc pipelines and institutional VMRs.
Apatosaurus was developed to address two specific gaps in existing infrastructure that render standard, cloud-based tools inadequate for certain types of philological work:
While Handwritten Text Recognition (HTR) and digital facsimiles are increasingly prevalent, many complex manuscripts remain undigitized, or their digitizations are insufficient for detailed codicological and paleographic analysis. Consequently, direct transcription in manuscript rooms remains a methodological necessity. However, cloud-based VMRs assume a constant connection to save work. In archives with unreliable networks, this risks severe data loss. An environment was needed that treats offline capability as the default state, rather than a graceful degradation.
Manuscripts such as those in Family 0150 are catenae, interleaving biblical text (lemmata) with patristic commentary (scholia), which often accounts for half the text on a given page (Flood 2024). Most transcription tools treat text as a linear sequence and lack mechanisms to intuitively represent the spatial and logical relationships between a verse and its surrounding commentary without requiring the manual authoring of complex TEI XML.
Apatosaurus implements a "Local-First" architecture (Kleppmann et al. 2019). The core principle dictates that the user's browser holds the primary copy of all data; the server, when available, functions merely as a synchronization relay.
Built as a Django and SvelteKit application utilizing a custom sync engine and WASM SQLite for local persistence, the tool relies on a Service Worker to cache the entire application upon initial load. As a PWA, it runs on any modern browser—including mobile devices—without requiring platform-specific installation. Crucially, all parsing logic executes client-side. A TypeScript parser operating in the browser ingests TEI, tokenizes readings, and renders the interactive transcription locally.
This architecture serves as its own preservation strategy. Digital humanities tools frequently become unavailable when grant funding ends and servers are decommissioned (Smithies 2017). Because Apatosaurus runs entirely client-side, the tool remains functional even if host domains expire. To mitigate the risk of users accidentally clearing browser caches (OPFS), the application includes streamlined mechanisms for "ejecting" data, allowing users to export their work directly to standard TEI XML or JSON formats at any time.
In addition to offering a robust and fully featured transcription editor, Apatosaurus enables the collation of these transcriptions for the construction of a digital critical apparatus. One of the challenges with the collation of handwritten manuscripts is that there are often layers of corrections. This can mean that there are many “witnesses” to a text within the same transcription. Apatosaurus enables corrector’s texts to be treated as either fragmentary or full. By fragmentary, I mean that we allow a corrector’s text to be present and collated only where there are explicit corrections by that hand. By full, we assume a corrector agrees with the first hand except where corrections occur. The user of Apatosaurus can choose per witness per collation how to handle these cases. The computation of extracting the same section of text from many transcriptions, handling additional scribal hands per transcription, and the ultimate collation is all handled on the user’s device entirely in the browser.
The architecture of Apatosaurus reflects the specific constraints of field research: operating offline, handling non-linear texts, and enabling low-friction collaboration without dedicated IT support. While not intended to replace institutional infrastructure for large-scale, heavily funded projects, it provides an alternative paradigm for independent scholars and small teams. By treating offline work as the norm and utilizing the client as the primary computational environment, local-first tools offer a highly sustainable model for digital philology. (Screenshots and live demonstrations of the interface will be provided during the presentation).
The software is open-source and freely available under an MIT license at https://github.com/d-flood/Apatosaurus