Skip to main content
  • Status: accepted · Date: 2026-07-17 · Deciders: Davor Runje

Context

Should the literature registry share the dataset manifest format? A field-level comparison showed literature and dataset registries are more different than similar — and literature has a mature standard format that datasets lack. The author also wants to record decisions about papers (interesting? act on?).

Decision drivers

  • Bibliography has a mature ecosystem standard (BibTeX/CSL-JSON, Zotero, pandoc).
  • The author needs a triage/decision layer (roles, dispositions, rationale) a bib file doesn’t have.
  • Provenance spine + mirror mechanism are shared with datasets (ADR-0012).

Considered options

  1. Standard bib (facts) + a git-tracked triage sidecar (decisions), joined by citekey/DOI.
  2. A unified datasets.yml-style manifest for both papers and data.
  3. Bib only (no triage).

Decision

Option 1. Bib = immutable facts (BibTeX/CSL-JSON); triage YAML = our decisions (role, disposition state-machine, rationale, seeded links, intent). acted-on → backlog; screened + rationale → PRISMA log.

Consequences

  • Reuses the citation ecosystem; captures the decision trail the author wanted.
  • Two files per corpus (bib + sidecar), joined by key.

Rejected alternatives

  • Unified manifest — throws away the bib ecosystem; forces a hand-rolled clone.
  • Bib only — no place for triage decisions / PRISMA reasons.
sub-spec 2 §4; meta-spec §3.4.