- Status: accepted · Date: 2026-07-17 · Deciders: Davor Runje
Context
Where does the authoritative state live — external services (Zotero DB, MLflow, W&B) or git-tracked files? And should a thesis be able to aggregate papers across separate repos?Decision drivers
- Reproducibility, diffability, PR-review, and shareability with peers without a shared account/service.
- Consistency with mononet’s committed-provenance stance.
- Work-research and PhD thesis live in separate repos with separate lives.
Considered options
- Git-tracked plain text (bib/YAML/markdown + committed provenance) is the source of truth; external tools are optional front-ends/caches.
- External trackers (Zotero DB, MLflow/W&B) as source of truth.
- Support cross-repo thesis aggregation now.
Decision
Option 1. Zotero is an optional bib front-end; Optuna/.doit.db and rebuilt
caches are not authoritative; registries/decisions/results are committed.
Cross-repo aggregation is explicitly out of scope for now (each repo’s top
level is self-contained) — recorded as a future item / GitHub issue.
Consequences
- Everything is reviewable and reproducible from the repo alone.
- A future cross-repo thesis roll-up would need a separate design.
Rejected alternatives
- External trackers as source of truth — not git-native; heavy; provenance not repo-owned; needs shared accounts.
- Cross-repo now — out of scope; adds coupling before it’s needed.