This month, the World Bank’s Reproducible Research Repository (RRR) published its 500th reproducibility package—a milestone for open science and research transparency. Until a few years ago, finding the data and code behind a World Bank publication could be difficult: data citations were scarce, and code files, if published at all, were scattered across journal repositories, GitHub, and personal websites. RRR was created to change that by providing a single home for reproducibility packages associated with World Bank Group publications. As we mark this milestone, we’re reflecting on what the first 500 packages have taught us about making research more transparent, accessible, and reproducible in practice.
The Reproducible Research Repository is the public-facing platform of the World Bank’s Reproducible Research Initiative, which builds on the World Bank’s broader commitment to open data and open knowledge. The initiative began in 2023 with a pilot for the Policy Research Working Paper (PRWP) series (described here). In 2025, that pilot scaled to a reproducibility requirement for PRWPs and a strong encouragement for other World Bank publications (described here). Today, RRR includes collections of Policy Research Working Papers, journal articles authored by World Bank staff, and flagships and other major analytical products.
Every package that’s published to RRR is verified for ‘computational reproducibility’ before it goes live. Our team checks that the package is complete and that the outputs exactly match the publication, and we work with authors to fix any issues we find. Publications that pass verification receive a Reproducible Research seal linking them directly to their reproducibility package on RRR (examples: Working Paper, indicator).
Beyond hosting data and code, RRR is designed to maximize the discoverability and usability of the packages. Each entry includes metadata on the software used, approximate run time, operating system requirements, and whether underlying data are openly included, publicly accessible but not redistributable, or restricted. Our team also checks that all data sources are properly cited and tags packages by data access, making it easy for users to identify packages they can inspect, run, and build on.
The Numbers Behind 500 Packages
Our first 500 packages are primarily Policy Research Working Papers (82%), which was the initial focus of the reproducibility initiative. They also include packages from our growing collection of flagship products and other WBG reports (12%), as well as journal articles by World Bank authors (6%). In the process of publishing these 500 packages, we’ve documented 2,476 unique datasets. Over 40% of the packages use fully open data, accessible to all users. Those that rely on proprietary or confidential data offer full documentation of the data sources, with access instructions, and include the code.
Data Access Type
Figure 1: Data Access Status for RRR Packages
In terms of software, 4 out of 5 packages use Stata, with R a distant second, and Python third. Other software such as MATLAB, EViews, and ArcGIS are represented in smaller numbers.
Figure 2: Software(s) used in reproducibility packages on RRR
Note: A package may use multiple languages; bars show total occurrences.
While some packages are highly complex, most (73%) can be run in under an hour – making them useful as practical examples for teachers and students.
Figure 3: Computational Time (as reported on RRR)
The 500 packages span 76 countries, with the largest concentration in Sub-Saharan Africa (106 of the 500 packages). The most frequently represented countries are Ethiopia, Brazil, Bangladesh, India, Nigeria, and Pakistan. 196 of the packages are for global analytical products.
The repository’s audience extends well beyond the Bank’s own researchers. RRR has attracted more than 80,000 unique visitors from 184 countries, suggesting broad interest in accessing verified code, data documentation, and research workflows. Nearly 20,000 users have downloaded a reproducibility package. The most viewed and downloaded packages include widely used analytical products such as the Worldwide Governance Indicators, Business Ready, and the 2025 World Development Report, showing that demand is especially high for resources that help users understand and reuse flagship outputs.
The Journey to 500
RRR is growing quickly. It took nearly three years to reach the first 500 packages, but we expect the next 500 to arrive much faster. The publication rate has increased each year since RRR launched. Today, we publish about 25 packages per month, a 65% increase over the previous year. This acceleration reflects both stronger institutional uptake and growing familiarity with reproducibility requirements.
Figure 4: The Journey to 500
Lessons from verifying 500 packages
Scaling does not mean compromising on quality. One lesson from the pilot is clear: verification is essential for the credibility of the repository. Only one in five submitted reproducibility packages is computationally reproducible on first review – meaning our team is able to run the data and code provided and obtain the exact results reported in the publication. The average package goes through more than two rounds of revision before it is ready for publication on RRR.
This is not surprising. A growing body of evidence shows that data and code availability policies need active reinforcement to be effective. Reproducible research standards and workflows are relatively new within the World Bank, and many staff members have not previously had to think about whether their code can run on a different machine -- without their personal file paths, local package installations, or undocumented manual steps. To address those challenges we’ve developed a set of open-source tools, including guidance for authors, an interactive README generator, tools and templates, environment setup guides for R, Python, and Julia, and the repkit package for Stata, which handles portable file paths, package version locking, and output stability checks.
Three years in, norms and practices are beginning to shift. We see encouraging evidence of learning by doing. Packages submitted by authors who have already gone through the verification process are more likely to reproduce on first try and are faster to review, suggesting that reproducibility is a learnable practice that becomes easier to integrate over time.
Figure 5: Package Quality and Reviewer Hours by Experience Tier
Why It Matters for Global Development Research
The RRR is not only a transparency tool — it is also a practical learning resource. Users can browse packages by software, topic, or data accessibility and find, in a few clicks, a complete, verified example of how an applied development economics paper is actually built. The repository is also a ready-made catalogue of potential replication projects, especially for young researchers looking for examples that run on open or publicly accessible data. RRR offers great opportunities to inspect, learn from, reuse, and extend World Bank research. Start browsing: five hundred packages are waiting, and many more are on the way.
Join the Conversation