Granular Provenance Metadata v0.2 extends the provenance schema below the semantic layer.
An eighth type—spatial-typographic mediation—records whether spacing, lineation, typography, prosodic notation, stanza groups, manuscript marks, or page geometry carry meaning and what happened to them during machine preparation.
The new representation pipeline traces the work from source artifact through digitization, normalization, serialization, tokenization, and model access. Loss can occur at every stage. A non-OCR scan may never become operable text. A calligram may preserve every phrase but lose its spatial argument during serialization. A transcription may retain stanzas while normalizing manuscript dashes.
Four statuses describe the archive state: compositionally invisible, compositionally reduced, compositionally faithful, and compositionally operational. The last means the archive’s compiler can address the work’s composition, not merely its propositions.
The version also formalizes repository versioning. Every substantive version receives a distinct deposit. Successors declare ancestry. Predecessors acquire superseded_by through registry-only edits. Latest-version resolution follows the chain. Correction-in-place is separated from version creation and reserved for defects that prevented the intended bytes or metadata from being represented.