Note / 02 / Reproducible research
Designing research that can be re-run
What changes when the manuscript, dataset, and evaluation code become one inspectable object.
- Project
- RAVA
- Format
- Research field note
- Reading time
- 4 min read
The gap between reading and verifying
A paper can describe an experiment clearly and still make verification difficult. The manuscript may live in one place, the dataset in another, and the evaluation code somewhere else entirely.
That friction matters most when a result depends on choices that are easy to compress in prose: filtering, aggregation, train-test construction, thresholds, and evaluation metrics.
One publication, three connected parts
RAVA stands for Reproducible, Auditable, Verifiable Articles. The format treats a manuscript, its dataset, and its evaluation code as one interactive publication rather than three loosely connected attachments.
The goal is not to guarantee that every result is correct. It is to make the audit path available: move from a statement to the relevant evidence, inspect the choices, and rerun the evaluation when a question deserves a closer look.
Publishing is also interface design
An auditable article needs to show where evidence begins, which parts are interpretation, and how an output was produced. That makes information architecture part of research communication.
The first RAVA implementation uses the Sysmon benchmark study as a working example. It does not replace the manuscript; it gives the manuscript an executable evidence layer.
The direction from here
The larger idea is simple: confidence should not depend on the author's summary alone. A reader should be able to inspect the chain from claim to artifact without rebuilding the project from scattered links.
RAVA treats that connection as a design constraint: a research object becomes more useful when its claims can be traced cleanly to the evidence behind them.
Working takeaways
What stays with the project.
- 01Keep the manuscript, dataset, and evaluation code together.
- 02Design a visible path from each claim to its evidence.
- 03Treat reproducibility as part of the publication, not an appendix.