- Status: accepted · Date: 2026-07-17 · Deciders: Davor Runje
Context
literature and dataset both manage externally-sourced, identified, licensed,
mirror-able assets. Are their registries the same? A field-level comparison found
a modest shared provenance spine but larger, non-overlapping distinctive fields —
and literature has a standard format datasets lack (ADR-0008).
Decision drivers
- Avoid a redundant second mirror/fixity/ID implementation.
- Don’t force two genuinely different registries into one file.
Considered options
- Share the mechanism (asset spine + rclone mirror + fixity + persistent-ID), keep two distinct registry files/formats.
- Fully unify into one registry.
- Fully independent literature and dataset stacks.
Decision
Option 1. One base asset record + one materialization/mirror engine in the substrate; literature (bib+triage) and dataset (datasets.yml) extend it.
Consequences
- No duplicated mirror logic; each front-end keeps its natural format.
- A base-record schema both must conform to.
Rejected alternatives
- Full unify — false equivalence; loses the bib ecosystem.
- Fully independent — duplicates the non-trivial mirror/fixity/ID machinery.