Abstract

Fragmentary inscriptions are often approached through corpus similarity ranking, in which candidate restorations are prioritized by textual proximity. Under fragmentary conditions, similarity ranking can introduce interpretive assumptions before physical and structural constraints have been fully resolved. This study examines what changes when structural constraint elimination precedes similarity ranking in fragmentary inscription analysis.

Using Agora XV 354 (I 7450) as a bounded test case, the analysis proceeds from broad structural compatibility toward progressively narrower interpretive space through explicitly defined constraint domains. Material constraints (legibility modeling, letterform stability, margin preservation), structural constraints (segmentation behavior, clause termination, and formulaic compatibility), and distributional constraints (corpus frequency comparison, adjacency testing, and absence-based constraint testing) are applied in ordered phases. Interpretations that violate physical, syntactic, or distributional boundaries are removed before any similarity ranking is introduced.

Similarity comparison is engaged only after the candidate space has been structurally constrained, reducing assumption load prior to interpretation. The resulting constrained configuration was subsequently evaluated against an independent probability-first masked restoration model (Ithaca), which produced convergent structural segmentation without contradiction.

The case does not assert definitive reconstruction or reclassification. Rather, it demonstrates how explicit sequencing of constraint domains can provide a reproducible and assumption-minimizing framework for evaluating fragmentary inscriptions.

Back to Day 1 programme