Posts
Restored, and Marked Wrong
The 2025 federal dataset removals, the librarians who archived 311,000 datasets before they could fully vanish, and the record that came back carrying a disclaimer against itself
The source of this silence is a single, traceable instruction. On January 20, 2025, his first day back in office, the President signed Executive Order 14168, directing federal agencies to combat what the order termed "gender ideology" and to use the word "sex" rather than "gender" in federal policy and documentation. Nine days later, on January 29, the Acting Director of the Office of Personnel Management issued a memorandum requiring agency heads to terminate any programs and remove any outward-facing media that "promote or inculcate gender ideology."
That memorandum is the documented hinge of the entire post. It is dated, attributed, and a matter of public record — there is no ambiguity about where the instruction originated or when it issued. What followed was not a single agency's discretionary judgment call. It was a coordinated compliance cascade across the federal government, beginning within forty-eight hours.
The conduit is the agency-level compliance mechanism that converted one memo into thousands of individual removals. In the weeks that followed, government organizations removed or modified over 8,000 web pages and approximately 3,000 datasets, concentrated in content related to diversity, equity, and inclusion initiatives, gender identity, public health research, environmental policy, and related programs. The Centers for Disease Control and Prevention saw over 3,000 pages altered or removed; the Census Bureau removed roughly 3,000 pages of research materials. Major datasets disappeared from the Department of Energy, the National Oceanic and Atmospheric Administration, and the Environmental Protection Agency.
Harvard Law School Library's own before-and-after snapshots of data.gov give the conduit a precision the series hasn't had before: 307,854 datasets were listed on January 21; by the end of that month, more than 2,000 had disappeared. This is the fastest-moving silence the series has documented — a gap measured in days, not decades.
The conversion mechanism in this specimen operates two ways, and the second is the one that makes it distinct from every prior post in the series. The first conversion is the familiar one: existing public data converted into absence — a page or dataset simply taken down, the same basic mechanism as a redaction, just executed by removal rather than blackout. The second conversion is new to the series: existing public data converted into contested data, where the content itself survives but is paired, on the same page, with language asserting the content is wrong.
That second mechanism is harder to categorize than outright removal, and that difficulty is the point. A removed page is an absence anyone can notice and measure — which is exactly what allowed Harvard's snapshot comparisons and the Doctors for America lawsuit to document it precisely. A restored page carrying a disclaimer against itself is neither fully present nor fully absent. It exists, technically satisfies a restoration order, and still performs the original removal's underlying argument — that the information should not be trusted — through a different mechanism than deletion.
Jack Cushman, director of Harvard's Library Innovation Lab, framed the project not as a partisan response but as a continuation of a function libraries have performed for centuries — drawing an explicit comparison to the library's thirty-nine early manuscript copies of the Magna Carta. The shadow archive in this specimen did not present itself as resistance. It presented itself as ordinary institutional memory, doing what libraries are for.
The archive's technical design matters as much as its existence: datasets were packaged using BagIt, a Library of Congress preservation standard, with cryptographic signatures and checksums attached specifically so the copies could be verified as authentic decades from now — the same evidentiary rigor a court would expect of a primary source.
Litigation outcomes complicate any simple resolution. The court that ordered the February restoration also found, in its July 2025 ruling, that the plaintiffs lacked standing to challenge the disclaimer language specifically — meaning the disclaimer mechanism survived judicial review even after the underlying removal did not. Restoration and correction turned out to be two separable legal questions, and only one of them was actually won.
A restored page carrying a disclaimer against itself is neither fully present nor fully absent.
The Silence Architecture · Series AnalysisThe insulation in this specimen is procedural in the same register as Post III's certifications, but compressed to a fraction of the timescale. Judge Bates's opinion found that the agencies likely violated the Administrative Procedure Act and Paperwork Reduction Act specifically because they removed the pages without notice or explanation — the insulating failure was not secrecy about what had been removed, but the absence of any stated justification for removing it, which is itself a procedural requirement the agencies are obligated to satisfy and did not.
The disclaimer is the more durable insulation, and it is the layer that survived the litigation. By the time the case reached final judgment in July, the agencies' own compliance report concluded their review had not identified any further pages legally required to be restored — meaning the matter closed not with full resolution but with the government's own determination that it had done enough. The disclaimer language, found by the court to be outside the plaintiffs' standing to contest, remained in place on at least some restored pages even after the underlying removal had been ruled likely unlawful. Insulation here did not require hiding the record. It required only attaching enough doubt to the record that restoration and retraction could coexist on the same page.
This is the specimen's contribution to the series, and it completes a pattern worth naming across all five posts now. Post I needed sixty years and a social movement. Post II needed eighty-three years and a new statute. Post III needed thirty-three years and a single presidential order. Post IV needed 104 years and arrived at acknowledgment without remedy. This post needed three weeks for removal, three weeks more for partial court-ordered restoration, and five months for a final ruling that left part of the mechanism intact. The architecture is the same in every case. Only the clock has changed — and a faster clock does not yet mean a more complete one.
Two of three conditions fire. This post earns the tag.
Temporal Capital — fires at a different scale than any prior post in the series. The lag between removal and partial restoration is dateable to the day: memo on January 29, lawsuit filed February 4, restraining order February 11, final summary judgment July 3. The compressed timescale is itself diagnostic — it shows the same friction mechanism operating under modern litigation speed rather than historical inertia, and demonstrates that fast resolution does not automatically mean complete resolution.
Interpretive Capital — fires through the disclaimer mechanism specifically. The same underlying medical and scientific content was alternately presented as authoritative federal guidance, then absent, then present again but explicitly labeled as incorrect — three distinct interpretive states applied to one unchanged body of data within a matter of weeks.
Enforcement Asymmetry — does not clearly apply. The OPM memo was a uniform directive applied consistently across agencies, not a constant rule enforced unevenly between similarly situated parties. Per the v5.5 standard, this condition is not forced into the read.
Who absorbed the redistributed friction: the physicians, researchers, and patients named directly in Judge Bates's opinion — particularly the clinicians treating low-income and underprivileged populations who lost real-time access to clinical guidance — bore the immediate cost, while the agencies that removed the data without the required notice faced no consequence beyond a court order to restore some, but not all, of what they had taken down.
Executive Order 14168 and the January 29, 2025 OPM memorandum are matters of public federal record. The dataset and webpage removal figures — approximately 3,000 datasets and over 8,000 web pages, with agency-specific figures for the CDC and Census Bureau — are drawn from the Wikipedia account of the 2025 federal resource removals, which is itself sourced to contemporaneous reporting; this post treats that secondary aggregation as Tier 2 evidence rather than upgrading it to Tier 1. The data.gov snapshot counts (307,854 datasets on January 21, more than 2,000 disappeared within ten days) are drawn from 404 Media's contemporaneous reporting via Harvard researcher Jack Cushman's own before-and-after archive comparison, a Tier 1 primary observation. The Harvard Law School Library Innovation Lab's 311,000-dataset, 16-to-17-terabyte archive, its BagIt preservation methodology, and Jack Cushman's statements are drawn directly from the Library Innovation Lab's own published announcements and from contemporaneous reporting in the Harvard Crimson and Library Journal. The Doctors for America v. OPM litigation history — the February 4 filing, the February 11 temporary restraining order, the specific disclaimer language, and the July 3, 2025 mixed summary judgment outcome including the standing finding on the disclaimer claim — is drawn from the Civil Rights Litigation Clearinghouse's case docket, a Tier 1 primary legal source, cross-checked against contemporaneous court reporting from NPR, CBS News, and FedScoop.
The series methodological note carries forward, with one addition specific to this post: this is the first specimen in the series describing events still close enough to the present that the full legal and administrative record may continue to develop after publication. The Wall requires that this be stated directly rather than presented as a closed case. As of this writing, the underlying executive order remains in effect and the agencies' compliance posture reflects their own July 2025 determination that no further restoration was legally required — a determination this post reports, but does not treat as the final word on the matter.
No comments:
Post a Comment