Code and data transparency
When I joined the Cook County Assessor’s Office in 2018, the agency was emerging from a property assessment system that had lost public credibility. The new administration had promised fair, ethical, and transparent assessments. For the Data Department, transparency could not mean publishing a methodology memo after decisions had already been made. If software was going to determine the assessed value of more than a million properties, the public needed to be able to inspect the software, the data, and the reasoning behind it.
When policy is implemented through software, code is policy.
The problem
Property assessments distribute the cost of local government across homeowners, businesses, and communities. Small technical decisions can shift substantial tax burdens from one group to another. Yet the models and workflows behind those decisions have traditionally been difficult—or impossible—for the public to examine.
That opacity creates two problems. It limits accountability when the system produces inequitable results, and it prevents outside experts from helping the agency identify errors and improve its methods.
We needed to modernize the valuation system while also making it more open than the system it replaced.
What I did
As the office’s inaugural Chief Data Officer, I built and led a Data Department that treated transparency as a feature of production work rather than a separate publication process.
We established a public-by-default operating model:
- Production valuation code was published in public repositories.
- The data needed to understand and evaluate assessments was released through Cook County’s open-data platform.
- Model documentation, data definitions, lineage, and internal working practices were made publicly available.
- Analytical work was designed to be reproducible, including the release of scripts and underlying data rather than only finished charts or conclusions.
- Issues, mistakes, and methodological limitations were documented instead of hidden.
The goal was not to claim that the models were perfect. The goal was to make the system inspectable enough that taxpayers, journalists, researchers, and practitioners could determine where it worked, where it failed, and how it could improve.
Building transparency into the institution
Publishing code once would have produced a demonstration project. I wanted transparency to survive as an institutional practice. The Data Department’s mission, vision, and values therefore made transparency explicit. Our handbook stated that anything used in production should be available to the public unless there was a documented legal, privacy, or security reason to restrict it. It also defined reproducibility as providing the raw data, scripts, and documented steps needed for another person to reach—or challenge—our conclusions.
This changed how the team worked internally. Public documentation improved institutional memory. Version control made changes traceable. Open issues exposed unresolved problems. External review became part of the quality-control process rather than a threat to it.Transparency also imposed discipline. A team that expects its work to be examined has stronger incentives to document assumptions, explain decisions, and confront mistakes directly.
The result
The office published the production models used to value residential and condominium properties, along with data infrastructure, documentation, open datasets, and tools for analyzing the property-tax system. This made a consequential government system materially more auditable. It allowed outside experts to move beyond debating the office’s intentions and examine its actual methods. It also created a reusable model for public-sector data science: government agencies can operate modern analytical systems without requiring the public to accept those systems on trust.
The work continued after my departure. The CCAO Data Department’s public GitHub organization now contains its valuation models, data architecture, documentation, analytical tools, and public engagement materials.
What I learned
Transparency is easiest when nothing important is at stake. It becomes a real institutional value only when disclosure may expose an error, invite criticism, or complicate a decision. Public agencies should not publish code because code is inherently understandable or because open source guarantees good policy. Most residents will never read a valuation model. But independent experts can, and their ability to inspect the system changes the balance of accountability.
Transparency does not eliminate mistakes. It makes mistakes easier to discover, harder to conceal, and more likely to produce institutional learning.