Fragment-level FAIRness: annotating scientific data and its provenance using data fragment selectors

Scientific communication depends on the production of FAIR (findable, accessible, interoperable, reusable) data. Yet when datasets are shared, they remain hard to interpret across domains because annotations are tied to heterogeneous file formats and implicit semantics. We present a format-agnostic method for fragment-level annotation based on Data Fragment Selectors (DFS): URI-addressable pointers to sub-file regions (e.g., CSV rows/columns, JSON objects) whose meaning is defined by an accompanying explicit selector standard. DFS are aggregated into a Datamap, a lightweight schema that binds referenced fragments to ontology-backed semantics, including units and object types. Separating structural addressing from semantic interpretation enables reusable schemas and robust automation without format-specific parsing. To support provenance annotation, we integrate DFS with the Investigation-Study-Assay (ISA) metadata model, extending data-entity references to enable sub-file granularity. We realize the model in two interoperable serializations: ISA-XLSX for spreadsheet-centric practice and an Research Object Crate (RO-Crate) Datamap profile for web-native packaging and exchange. These specifications deliver fragment-level FAIRness: precise, machine-actionable links between process metadata and the data portions they describe. We demonstrate practical applicability by introducing a Spreadsheet Fragment Identification specification, providing a standardized workflow for annotating aggregated supplementary data in journal publications, and via programmatic exploration of a real-world proteomics project.