On March 6, 2020, two reporters at The Atlantic, Robinson Meyer and Alexis Madrigal, published the number of Americans they could confirm had been tested for the new coronavirus: 1,895. To get it they had contacted the health department of every state and the District of Columbia and read whatever each had posted. Through February the Centers for Disease Control and Prevention had published its own running tally; the figure on February 29 was 472. On March 3 the agency announced it would stop publishing the number, and when it resumed two days later it counted only the tests the agency itself had completed. The day after the story ran, the reporters' tracker merged with a spreadsheet Jeff Hammerbacher had been keeping on his own, and the COVID Tracking Project put out a call for volunteers.
Collecting the numbers turned out to be harder than copying them. Some states built dashboards; others gave figures at press conferences or not at all. Some reported specimens and some reported people, and in those first weeks most labs needed two specimens to test one person, so Colorado's capacity of 160 specimens a day meant about 80 patients. Early on, as many as ten states in a single day might overhaul their sites or launch new dashboards. Madrigal later told CNN that the team had expected to pull numbers off dashboards and found instead that it needed to do deep research.
So the work stayed manual. Every day, volunteers entered and double-checked roughly 800 figures from 56 jurisdictions, and a screenshot system eventually captured the source page behind every one of the project's 797 core data points, so any number could be traced to what a state had shown that day. Engineers wrote scripts to check the volunteers' entries, but the project never fully automated a single data point, because a person copying a figure also read the note printed beside it. In November 2020, the project noticed that Hawaii was reporting 16,096 cumulative hospitalizations on one dashboard and 1,125 on another. It flagged the gap to the state's emergency management agency, and Hawaii fixed its arithmetic: 1,125 was right. The project estimated more than 20,000 hours of data entry alone.
Researchers used the data in nearly 800 academic papers, and two presidential administrations cited it in strategic plans. The project kept working relationships with health departments in states governed by Republicans and by Democrats. On February 1, 2021, Madrigal and Erin Kissane, the project's managing editor, announced that it would stop. Compiling and standardizing data from 56 jurisdictions, they wrote, was properly the work of federal public health agencies, and the CDC and the Department of Health and Human Services now looked able and willing to do it. The last daily update went out on March 7, 2021, a year to the day after the project began.
A count of tests or hospitalizations looks like the plainest kind of fact. It is only as good as the agreement about what is being counted and the attention of whoever reads the footnote, and for a year much of that attention came from volunteers checking one another's entries against screenshots. They stopped when the job could go back where it belonged. A country cannot answer an outbreak it has stopped counting.