The 2026 Census Test no longer tests most of what it was built to test

The Census Bureau is in the middle of a 14-year life cycle (2019-2033) preparing for the 2030 Census, and its main tool for validating new design ideas before locking them in is the 2026 Census Test.▸ A June 2026 GAO report⧉ found the Bureau cut that test from 6 planned sites to 2, and from 19 operational activities it set out to test for viability to 9 -- while total agency staff fell 16 percent, from 13,312 to 11,119, in nine months. A second June 2026 GAO report⧉ examined a different piece of the same preparation: the Enterprise Data Lake, one of four IT systems the Bureau is building to collect, store, and process 2030 Census data. Its schedule, GAO found, does not substantially meet any of the four characteristics GAO uses to judge whether a program's completion dates can be trusted.

The 2026 Test's stated purpose was to determine whether the Bureau should keep pursuing a list of enhancements it had identified through early research -- things like continuously updating the address list between censuses, processing data in near-real time, and simplifying field operations.▸ To find out, the Bureau planned to run the test at six sites and check 19 separate operational activities for whether they actually worked. Instead, four of the six sites were dropped because they weren't part of a separate U.S. Postal Service pilot the Bureau decided to prioritize, and 10 of the 19 activities were reduced in scope or removed from testing altogether. The remaining two sites also saw internet self-response delayed 50 days and in-field enumeration delayed more than two months, and the test's total hiring goal was cut by more than 90 percent.

The activities that got cut include some of the more technically ambitious ones: modeling how well in-office enumeration works, testing new ways to process addresses without an identifier, and a group-quarters internet self-response option.▸ GAO's own standards for evidence-based policymaking call for agencies to build a portfolio of credible evidence before finalizing a decision -- identifying what evidence is needed, generating it, and assessing whether what exists is sufficient. GAO pointed to the 2020 Census as a precedent for what happens when that step gets skipped: the Bureau ran into problems submitting group-quarters data electronically in 2020 after reducing its testing of that same feature during the run-up test in 2018.

The test that remains also asks more of respondents.▸ Its internet questionnaire includes new questions on citizenship, education, and housing, and takes about 40 minutes to complete online -- four times the roughly 10 minutes the 2020 Census questionnaire took. Commerce approved an English-only version as sufficient for the test's objectives, and the Bureau has no plans to hire bilingual enumerators for it, even though the 2020 Census fielded questionnaires in 12 non-English languages, including Spanish and Chinese.

The staffing picture compounds the testing gap.▸ Bureau officials attributed the 16 percent staff decline largely to early retirement and deferred-resignation programs taken amid a federal hiring freeze, and said the agency now has skills shortages in areas including field-operations modernization. The Bureau has launched an agencywide assessment to identify those gaps -- but as of February 2026, that assessment was paused until the Bureau finishes establishing roles for the employees who'd participate in it, a step officials said may not happen until an enterprise-wide reorganization tentatively scheduled for fiscal year 2027. That means the skills assessment could be finished after the Bureau has already hired and trained the workforce it's supposed to inform.

Separately from the field test, the Bureau is building four interconnected IT systems -- collectively called the Business Ecosystem -- to replace the modernization effort it built for the 2020 Census and then closed in March 2020 without finishing.▸ That earlier effort, CEDCaP, is itself a cautionary data point: its cost estimate grew from $3.41 billion in 2015 to $4.97 billion by 2017, driven partly by late decisions on IT capabilities and contractors, before the Bureau scaled it back and eventually shut it down having delivered only some individual systems, not the integrated capability it promised.

Of the four current programs -- data dissemination (CEDSCI, $753 million), data collection (DICE, $1.08 billion), data storage and analysis (the Enterprise Data Lake, $337 million), and a smaller data-linking research effort (Frames, about $12 million a year) -- GAO examined the Enterprise Data Lake in depth.▸ It found the program fully implements leading practices for managing risk and substantially implements practices for managing requirements and estimating costs; its cost estimate, in fact, was judged reliable. Its schedule was not.

GAO checks a schedule against four characteristics: comprehensive, well-constructed, credible, and controlled.▸ The Enterprise Data Lake's November 2024 schedule didn't substantially meet any of them. Seventy-two percent of its detailed activities had no assigned resources. More than a quarter of the remaining activities were missing the links to predecessor or successor tasks that let a schedule calculate dates at all, and 37 percent had date constraints GAO found unjustified -- locking activities to fixed dates regardless of how the work ahead of them was actually going. The scheduling software couldn't even produce a valid critical path, the sequence of activities that determines the earliest a program can finish; without one, GAO found, managers can't identify which slipping tasks would actually delay the whole program.

The four IT programs are meant to work as one system serving the same censuses and surveys, so GAO also checked whether the Bureau manages the dependencies between them.▸ It found partial implementation: the programs hold regular meetings to discuss status and risks, but there's no documented process for making sure that when a survey's onboarding date changes, the schedules of the IT programs that depend on it get updated to match. Without one, GAO found, schedule conflicts between programs could go unnoticed until they've already caused delays. GAO had flagged similar problems with CEDSCI, the dissemination-system program, back in April 2024; as of January 2026, the Bureau had drafted -- but not finalized -- the process documentation GAO asked for then.