Network analysis is often considered ill-suited to fragmentarily preserved texts. However, this paper argues that fragmentation should not be treated as a deficit but as an analytical object. By modeling the processes through which texts are produced, transformed, and editorially mediated, network analysis can reveal not only what survives but also how relational structures are reshaped.

This approach is illustrated by Damascius' vita Isidori (c. 496–526), a late antique philosophical biography preserved only in Byzantine excerpts. Rather than attempting to reconstruct the lost original, this study models the extant text as a layered relational system shaped by successive acts of selection and mediation. The analysis introduces an evidential scaling system (0–3) that classifies relationships according to textual certainty: Level 0 (editorial conjecture), Level 1 (contested attribution), Level 2 (probable attribution), and Level 3 (secure attribution). To capture the coherence of the relations themselves, the model also records whether relations occur within the same fragment or are reconstructed by combining separate fragments, producing five robustness classes that distinguish between attributional certainty and transmission coherence.

A quantitative analysis of 171 nodes and 300 edges reveals that 62.46% of relations are securely attested (Level 3), while 17.94% reflect modern editorial conjecture (Level 0). Strikingly 76.74% of all relations occur within single fragments (Robustness Classes A-B), indicating that most relations survive as locally coherent relational microstructures rather than emerging through editorial recombination. Although filtering reduces network density from 0.01 (Levels 0–3) to 0.006 (Level 3 only), the polycentric structure of the network nevertheless persists.

The study shows how evidential annotation and robustness classes enable network analysis to capture the epistemic structure of fragmentary traditions. Network analysis thus becomes a tool not for reconstructing lost social realities, but for analyzing how relational knowledge is preserved and transformed across time.

The circulation practices of verse have long been at the heart of early modern manuscript studies, with scholars studying not merely the movement of individual instances but cluster of poems, often copied out as a group that then successively travel together. Quantitative network analysis of existing datasets of early modern manuscript verse, collated and cleaned in the ERC-funded project 'STEMMA: Systems of Transmitting Early Modern Manuscript Verse, 1475-1700', offers a means to examine these theories at scale and a lens to identify manuscripts for closer textual study in ways analogue means have thus far been unable to achieve. Focusing on two poems - Henry Reynold's 'Black Maid to the Young Boy', and Henry King's response, 'The Young Boy's Reply' - this work uses network analysis to consider overlaps in manuscript miscellanies and maps the circulation of these poems in order to identify clusters of poems and potential shared materials, both building on and challenging traditional literature. This paper examines how network analysis models not only offer a means to identify traditional literary concepts, but re-evaluate their definitions as well, considering ontological questions around poetry in the act of translating literature into quantitative data and reflecting on how this process forces us to reckon with what traditionally stable literary concepts actually mean, or, at the least, demands a definitive answer to debates that often are whole fields in and of themselves. By using a small sample of early modern poetry as an example, it also seeks to offer a reusable model for other similar enquiries, and highlight the value of a small-scale model for future research.

We propose a method for quantifying translator intervention by using interlinear translation—a strict word-for-word gloss preserving source-language syntax—as a computational baseline representing a translational "Degree Zero." We define the *Intervention Vector* as the difference between a literary translation and its interlinear counterpart in a high-dimensional embedding space, quantifying both the extent and the direction of the translator's departure. We validate this approach on 74 New Testament translations across five European languages (English, French, Italian, Polish, and Spanish), each paired with a language-specific interlinear baseline. The magnitude of this vector recovers the theoretical translation spectrum without supervision, ranking translations from literal to paraphrase with statistically significant differences in the majority of pairwise comparisons. This unsupervised method applies uniformly across languages, requiring only an interlinear baseline for each target language.