CUF-Links: Continuous and Ubiquitous FAIRness Linkages for reproducible research

Ian T. Foster, Carl Kesselman

Computational research should, in theory, be perfectly reproducible. However, in practice, connecting a final result back to its original inputs remains a manual, error-prone challenge. Authors Ian Foster and Carl Kesselman describe this as a "socio-technical" problem where the "linkages" between data, code, and environment are lost during the long lifecycle of a research project. This paper proposes CUF-Links (Continuous and Ubiquitous FAIRness Linkages)—a set of six principles and lightweight tools designed to ensure that every step of scientific discovery is documented and discoverable from day one.

The Reproducibility Gap

Reproducibility fails because provenance—the chain of custody for data—is often broken. Scientists work across different tools, timeframes, and collaborators, frequently copying files and running scripts without recording exact versions. Existing tools often try to force scientists into "closed" environments (like a single Jupyter notebook), but real-world research is too complex for such constraints. The CUF-Links approach seeks to capture these linkages without disrupting existing workflows.

Traditional manual record-keeping with handwritten identifiers, illustrating the starting point for provenance tracking that CUF-Links seeks to digitize.

The authors propose a methodology based on six "CUF" principles:

  1. Identify everything (P1): Use lightweight, persistent identifiers (Minids) for every artifact, not just the final paper.
  2. Hold metadata close (P2): Use BDBags (based on the BagIt specification) to package data and metadata together so they aren't separated during transfer.
  3. Describe things consistently (P3): Use controlled vocabularies and data dictionaries to avoid "free text" chaos.
  4. Anticipate change (P4): Design systems that allow metadata schemas to evolve as the project grows.
  5. Never repeat (P5): Automate repeated tasks and assign identifiers to the code itself.
  6. Play it again—and again (P6): Regularly verify provenance queries, similar to "Continuous Integration" in software engineering.

Core Technologies: BDBags and Minids

The implementation relies on two core technologies. BDBags provide a hierarchical directory structure for big data. Notably, they support the "Holey Bag" concept: using a fetch.txt file, a bag can represent terabytes of data without containing the actual bits, instead providing links and checksums to fetch data securely via services like Globus. Minids are minimal, resolvable identifiers that are significantly easier to create than formal DOIs, making them suitable for intermediate research artifacts.

Real-World Applications

The paper demonstrates these methods through two diverse scientific applications:

  • Transcription Factor Binding Sites (TFBS): A multi-step pipeline that processed 2.4 TB of raw genomic data from the ENCODE project into a 350 GB final atlas.
  • Neuroscience Synapse Study: A six-year investigation into synapse formation in larval zebrafish. This study managed behavioral observations and high-resolution microscopy images via the Deriva platform.

Key Insights and Results

A central tenet of the paper is that FAIRness is a process, not an event. Researchers typically only worry about making data Findable, Accessible, Interoperable, and Reusable at the end of a project. CUF-Links argues it must be continuous and ubiquitous.

This approach yielded tangible results in the zebrafish study. By retaining "bad" data with proper metadata (P2), the team identified a specific microscope hardware malfunction that the vendor later confirmed. Furthermore, the authors reported that preparing data and creating DOIs for publication figures took minutes instead of days because the linkages were already in place. Reviewers of the resulting neuroscience paper specifically praised the "complete transparency" provided by the supplemental data bags.

Limitations and Future Work

Despite its successes, the authors acknowledge several challenges. Software volatility remains a concern; linking to a GitHub URL is insufficient if specific commit hashes or container images aren't recorded. Additionally, the methodology requires a cultural shift and "discipline" from researchers to adopt new habits. Finally, while CUF-Links addresses reproducibility (same data, same code), the harder problem of replicability (new data, same question) remains a target for future work.

Summary updated

Abstract

Despite much creative work on methods and tools, reproducibility -- the ability to repeat the computational steps used to obtain a research result -- remains elusive. One reason for these difficulties is that extant tools for capturing research processes do not align well with the rich working practices of scientists. We advocate here for simple mechanisms that can be integrated easily with current work practices to capture basic information about every data product consumed or produced in a project. We argue that by thus extending the scope of findable, accessible, interoperable, and reusable (FAIR) data in both time and space to enable the creation of a continuous chain of continuous and ubiquitous FAIRness linkages (CUF-Links) from inputs to outputs, such mechanisms can provide a strong foundation for documenting the provenance linkages that are essential to reproducible research. We give examples of mechanisms that can achieve these goals, and review how they have been applied in practice.
BibTeX
@article{foster2022cuf,
  author = {Ian T. Foster and Carl Kesselman},
  title = {{CUF-Links: Continuous and Ubiquitous FAIRness Linkages for reproducible research}},
  journal = {Computer},
  year = {2022},
  doi = {10.1109/mc.2022.3160876},
}
Figure