# Federal Disclosure Avoidance Crosswalk Every federal statistical agency publishes tables built from answers it promised to keep confidential, and every one of them does something to those tables first so that no single household, farm, business or patient can be read back out. The family of techniques is called **disclosure avoidance**, and there are five of them: suppression, coarsening, swapping, noise infusion, and synthetic data paired with a server that runs the real computation privately. They are not interchangeable. Each one protects a different thing, costs a different user something different, and — the question this map was built to answer — **is chosen by a different kind of instrument.** That last column is the one nobody has drawn, and it turns out to be the whole story. Sorting the system by *what instrument sets the method* sorts it by **who can change the rule**, and the answer is consistently: somebody well below the level of law. Congress wrote the promise and forbade the result. It never named a technique. The one place a method is written into binding federal text is a regulation about health records, written by an agency. The decision that actually governs the numbers in a census table is made by a departmental order that was never published in the *Federal Register* and by a committee whose minutes are not published at all. Part A is the five families and what each does to a number. Part B is the four tiers of authority, from statute down to internal board, with the operative words at each. Part C sorts the agencies by whose instrument can reach them. Part D is the review-board layer — who decides, body by body. Part E is the sharpest finding on the page: the line between what agencies publish and what they refuse to. Part F answers a question the method debate usually skips, which is *where on the ladder of aggregation each method actually bites*. Part G is the 2026 reversal and the three things it left unreconciled. Companion to [[Census Geography Tree]], which maps the ladder Part F runs down, and to [[Census Data Access Crosswalk]], which maps the routes a reader takes to the data these methods have already been applied to. For the one method this wiki already covers in depth, see [[Differential Privacy]]; for the order that ended its use at Commerce, [[Disclosure Avoidance for Statistical Products]]. For the agencies as a structure, [[Federal Statistical System Tree]] and [[Statistical Officials Crosswalk]]. > [!note] How to read this chart > **§** = the instrument doing the work, quoted at the text · **☐** = the rule is published as a number a reader can check · **⊘** = withheld on purpose, or stated nowhere at all · **◎** = the point where the choice of method is actually made, as distinct from where it is authorised · **≠** = two federal instruments that cannot both be obeyed · **★** = the structural payoff · **⚑** = changed in 2026 · **◈** = a convention nobody wrote down · **·** = separator for peer items on one line. > > Two glyphs are re-pointed for this subject and keep these meanings throughout: **☐** marks a published threshold rather than a published notice, and **◎** marks the body or instrument that picks the technique rather than a single official acting alone. > > **No individual is named anywhere inside the chart** — only offices, boards and instruments — so nothing in the structure decays when a seat turns over. Statutes and regulations are as in force November 2026. Agency practice is as each agency publishes it, read **6 November 2026**; where an agency's own page is out of date with respect to the 2026 order, Part G says so rather than silently correcting it. ``` FEDERAL DISCLOSURE AVOIDANCE CROSSWALK FIVE METHODS · FOUR TIERS OF AUTHORITY · AND ONE CHOICE NOBODY WROTE DOWN │ Legend: § the instrument, quoted ☐ the rule is published as a number │ ⊘ withheld, or stated nowhere ◎ where the method is actually chosen │ ≠ two instruments that cannot both be obeyed ★ structural payoff │ ⚑ changed in 2026 ◈ a convention nobody wrote down │ Time-slice: law as in force Nov 2026 · agency practice as published, read 6 Nov 2026 │ ■ PART A — THE FIVE FAMILIES, AND WHAT EACH ONE DOES TO A PUBLISHED NUMBER │ ├─◆ 1. SUPPRESSION — the number is withheld and a flag is printed in its place │ │ does ..... removes the cell, then removes further "complementary" cells so │ │ the first cannot be recovered by subtraction │ │ costs .... the small place and the small industry, which is where the cell │ │ was most likely to be sensitive and most likely to be wanted │ │ used by .. BEA, all survey tables · BLS, the quarterly employment census · │ │ EIA volume data · NASS · the Economic Census · and, since June │ │ 2026, the Census Bureau as a stated last resort │ │ ★ ITS COST IS MEASURABLE AND LARGE: the quarterly employment census │ │ "suppresses over 60% of all possible cells" on the agency's own account │ ├─◆ 2. COARSENING — the number is reported less precisely │ │ does ..... widens the category: ages into ranges, dollars into bands, │ │ counts rounded, geographies combined │ │ costs .... anyone who needed the detail; nothing is flagged as missing, so │ │ the loss is invisible on the page │ │ used by .. NCES public-use files · NCHS microdata · NASS rounding rules · │ │ EIA recoding · ⚑ the Census Bureau's preferred method since 2026 │ │ ◈ COARSENING IS MOVEMENT UP THE AGGREGATION LADDER BY DEFINITION, which │ │ is why "which method" and "at which level" are the same question │ │ asked twice. Part F draws it. │ ├─◆ 3. SWAPPING — records are exchanged between units before tabulation │ │ does ..... moves a household or a firm from one unit to another, so totals │ │ survive and the composition of a small unit does not │ │ costs .... the within-unit breakdown at the lowest published level — which │ │ for a census is the race-by-ethnicity detail in small places │ │ used by .. the 1990, 2000 and 2010 censuses · NCHS lists it as available · │ │ the Census population-estimates page still gives it as the │ │ example, on a page revised 1 June 2026 │ │ ⚑ ⊘ NOW TREATED AS PROHIBITED AT COMMERCE: the Census Bureau reads the │ │ 2026 order's ban on noise infusion to "include disclosure avoidance │ │ methods incorporating record or value swapping" │ ├─◆ 4. NOISE INFUSION — a random quantity is added to the value in place │ │ does ..... perturbs the published figure; the table's shape is untouched │ │ costs .... the small count, where the noise is large relative to the number │ │ sub-family: FORMAL PRIVACY, in which the total amount of noise is │ │ accounted for across every release from the same data and the │ │ accounting itself is a published number │ │ used by .. the 2020 census, on the published formal method · the Census │ │ employment-dynamics programme, still · County Business Patterns │ │ since 2005 · BLS, moving toward formal privacy from the other │ │ direction │ │ ★ ⚑ THE ONLY METHOD WHOSE PARAMETER WAS EVER PUBLISHED, AND THE ONE │ │ METHOD NOW BANNED AT COMMERCE. Part E and Part G. │ └─◆ 5. SYNTHETIC DATA WITH A VALIDATION SERVER — the file is replaced, not treated │ does ..... publishes modelled records that correspond to no real person, │ then lets an approved researcher mail in code that agency staff │ run against the confidential file and release only the output │ costs .... time and eligibility rather than precision — weeks of queue, an │ approved project, and an agency reviewer between the researcher │ and every number they get back │ used by .. the linked survey-and-administrative-records file built with the │ tax and social-security agencies · the synthetic business database │ ⚑ ◈ SURVIVES THE 2026 BAN BY ITS ARCHITECTURE, NOT BY EXEMPTION: │ synthesis used as the final protection for a published product is read │ as prohibited noise infusion, but synthesis "remains available for │ imputation, small-area modeling, simulation, and validation-server code │ development" — and in this design the synthetic file is a drafting │ surface, while the released numbers come from a run on the real data │ └─ AND BESIDE ALL FIVE, THE NON-METHOD: RESTRICTED ACCESS. Publish nothing at that level and move the user into a secure enclave instead. Roughly three dozen such centres now operate; the network added sites in 2024, 2025 and 2026 while published detail was being cut back. ★ Access is being scaled up as publication is scaled down, and the two trends are almost never discussed as one policy. │ ■ PART B — WHAT INSTRUMENT SETS THE METHOD, IN FOUR TIERS │ read top to bottom: the authority falls at every step, and the specificity rises │ ├─◆ TIER 1 — STATUTE ........... forbids a RESULT. Names no method. Anywhere. │ │ § 13 U.S.C. § 9(a)(2), census — no officer may "make any publication │ │ whereby the data furnished by any particular establishment or individual │ │ under this title can be identified." Thirteen operative words. The │ │ whole of the command. │ │ § 44 U.S.C. § 3572(c)(1), the government-wide confidentiality statute — │ │ data "shall not be disclosed by an agency in identifiable form," which │ │ § 3561(7) defines as anything permitting identity "to be reasonably │ │ inferred by either direct or indirect means." Its one gesture at │ │ mechanism is § 3572(a)(3), "controlling access to, and uses made of" │ │ the information — access control, not treatment of the table. │ │ § 26 U.S.C. § 6108(c), the tax statistics authority, in a subsection │ │ actually headed "Anonymous form" — no publication may "be associated │ │ with, or otherwise identify, directly or indirectly, a particular │ │ taxpayer." ◈ The heading promises a method and the text delivers a │ │ prohibition, which is the pattern of the entire tier. │ │ § THE DELEGATIONS, STATED AS DUTIES: § 3563(a)(2) — each statistical unit │ │ "shall adopt policies, best practices, and appropriate procedures." │ │ § 3506(e)(3) — every agency shall "ensure that disclosure policies │ │ fully honor pledges of confidentiality," presupposing policies the │ │ statute never describes. 20 U.S.C. § 9573(c)(1)(A) — the education │ │ director "shall develop and enforce standards." │ │ § 7 U.S.C. § 2276(a)(2), agriculture, the one near-exception in the Code — │ │ information may not be disclosed "unless such information has been │ │ transformed into a statistical or aggregate form that does not allow │ │ the identification of the person who supplied particular information." │ │ Names the family. Sets no parameter: no cell size, no dominance rule. │ │ § 44 U.S.C. § 3582(b)(2)–(3) comes closest of all — it orders standards │ │ for "removing or obscuring information" so identity is "less likely to │ │ be reasonably inferred," and a "comprehensive risk assessment ... prior │ │ to any public release." A comparative standard and a process, and still │ │ no technique; and it hands the content to OMB. See Tier 2. │ │ ⊘ ★ NO PROVISION OF THE UNITED STATES CODE USES THE PHRASES "disclosure │ │ avoidance", "differential privacy", "noise infusion", "cell │ │ suppression" OR "data swapping". The entire technical vocabulary of │ │ the system — including the method applied to the 2020 census — exists │ │ nowhere in statute, and can therefore be changed without Congress. │ ├─◆ TIER 2 — CODIFIED REGULATION ... two of them, pointing opposite ways │ │ ☐ 45 C.F.R. § 164.514(b)(2), health records — THE ONLY METHOD WRITTEN │ │ INTO BINDING FEDERAL TEXT, and it is checkable line by line: strip │ │ eighteen listed identifiers; remove "all geographic subdivisions │ │ smaller than a State ... except for the initial three digits of a zip │ │ code" where that three-digit area "contains more than 20,000 people"; │ │ and aggregate every age over 89 "into a single category of age 90 or │ │ older." The alternative route, § 164.514(b)(1), substitutes an expert │ │ who finds "the risk is very small" and must document the analysis. │ │ ◈ Written by the health department. Congress wrote none of it. │ │ ◈ AND IT RUNS ON CENSUS OUTPUT: the 20,000-person test is measured │ │ "according to the current publicly available data from the Bureau of │ │ the Census," so a census publication decision moves a health-privacy │ │ rule in a different department. │ │ 5 C.F.R. part 1321 (89 FR 82453, in force 10 Dec 2024), OMB's rule under │ │ § 3563 — the parent agency must ensure the statistical unit "is │ │ permitted to determine the methods for conducting statistical │ │ activities," and § 1321.3(a) "supersedes any guidance or Statistical │ │ Policy Directives to the extent that they conflict." ◎ It assigns │ │ the choice of method. It does not make it. │ │ ⊘ ⚑ AND THE ONE RULE THAT WOULD HAVE SET GOVERNMENT-WIDE STANDARDS HAS │ │ NEVER ISSUED. The § 3582 regulation had a statutory deadline of │ │ 14 January 2020. On OMB's own current agenda it still reads "Proposed │ │ Rule Stage," last scheduled for February 2026, with no Federal │ │ Register citation. Six years and ten months past the deadline, and │ │ the vacuum it left is what Tier 4 filled. │ ├─◆ TIER 3 — GUIDANCE ......... recommends, catalogues, and declines to decide │ │ Statistical Policy Directive No. 2, Standard 7.2 (2006) — on release, │ │ agencies "must ensure strict compliance with any confidentiality │ │ pledge to the respondents and all applicable Federal legislation and │ │ regulations." No technique named; it points at the catalogue below. │ │ ◈ ★ AND THEN NINE WORDS THAT DO MORE WORK THAN ANYTHING ABOVE THEM — │ │ Guideline 7.2.2, entire: "Do not publicly reveal parameters │ │ associated with disclosure limitation rules." Part E is what twenty │ │ years of agencies reading that sentence actually produced. │ │ OMB's confidentiality-statute implementation guidance (72 FR 33362, │ │ 15 June 2007) — agencies must "Apply appropriate statistical │ │ disclosure limitation (SDL) techniques," and OMB told a commenter │ │ that listing specific techniques was "beyond the scope of this │ │ implementation guidance." The declination is on the record. │ │ Statistical Policy Working Paper 22, "Report on Statistical Disclosure │ │ Limitation Methodology" (1994, revised 2005) — the canonical federal │ │ catalogue of every technique in Part A. Its own series index states │ │ that such papers "are not OMB Statistical Policy Directives," while │ │ directives "serve as minimum requirements." It has since been │ │ "revised as the Data Protection Toolkit," also non-binding. │ │ ⚑ ◈ THE CATALOGUE RECOMMENDS WHAT COMMERCE BANNED. Working Paper 22 │ │ lists noise addition, data swapping and inference-valid synthetic │ │ data as mainstream recommended practice — the three families the │ │ 2026 order removed from the department's own bureaus. │ └─◆ TIER 4 — DEPARTMENTAL ORDER AND INTERNAL BOARD .. where it is actually set │ ⚑ ◎ DAO 216-26, "Disclosure Avoidance for Statistical Products," │ Department of Commerce, effective 4 June 2026, as recited in the │ Census Bureau's own disclosure-board policy: │ • "Noise infusion shall not be used for any statistical product │ disseminated by the Census Bureau." │ • "Coarsening shall be the preferred category of Disclosure │ Avoidance methods for all statistical products." │ • "Suppression shall be permitted as a last resort, only to be used │ when coarsening is prohibited by law." │ Reaches the Census Bureau and BEA. ◈ It is not a rule: no proposal, │ no comment period, no Federal Register publication. The department's │ own page for it was unreachable on every attempt in November 2026. │ ◎ THE CENSUS BUREAU'S DATA STEWARDSHIP EXECUTIVE POLICY COMMITTEE — the │ body that approves every parameter the Bureau applies, that set the │ 2020 census privacy budget, and that hears appeals from the │ disclosure board, where "The DSEP decision is final." Co-chaired by │ two deputy directors; a two-thirds quorum; standing biweekly │ meetings. ⊘ Its July 2026 charter never uses the phrase "disclosure │ avoidance," and its minutes and decisions are not published. │ ★ SO THE CHOICE IS MADE AT THE LOWEST TIER ON THIS CHART, AND IT IS THE │ ONLY TIER WHOSE INSTRUMENTS ARE NOT PUBLISHED. Every tier above it │ either forbids an outcome, assigns the decision to somebody else, or │ recommends without binding. │ ■ PART C — THE AGENCIES, SORTED BY WHOSE INSTRUMENT CAN REACH THEM │ the column that matters is not which method an agency uses but who can change it │ ├─◆ REACHED BY A COMMERCE DEPARTMENTAL ORDER — a Secretary can rewrite it │ │ Census Bureau · Bureau of Economic Analysis │ │ carve-outs: data shared among Census, BEA and BLS under Title 13 or │ │ 44 U.S.C. § 3576 · reimbursable products not built on protected data │ │ ⚑ BEA's 2023 published plan to replace suppression with noise infusion │ │ for its trade-in-services surveys is foreclosed by the order │ ├─◆ UNDER THE CONFIDENTIALITY STATUTE AND AN ORGANIC STATUTE ONLY │ │ no departmental order on method was found for any of these │ │ BLS — Labor · NCES — 20 U.S.C. § 9573 · NCHS — 42 U.S.C. § 242m · │ │ NASS and ERS — 7 U.S.C. § 2276 · EIA · the social-security research │ │ office · the justice and transportation statistics agencies │ │ ★ ⚑ BLS SITS IN LABOR, SO THE COMMERCE ORDER DOES NOT REACH IT — and it │ │ is the one agency still moving TOWARD formal privacy while Commerce │ │ moves away from it. Two federal statistical agencies are now headed │ │ in opposite directions on the same question, for no reason internal │ │ to the question. │ ├─◆ UNDER THE TAX CODE — and holding a veto over somebody else │ │ the tax statistics office, under 26 U.S.C. §§ 6103 and 6108 │ │ ◈ THE GATE RUNS OUTWARD: Census policy requires that "all microdata │ │ releases that are derived from Title 26 data are reviewed by the" │ │ tax agency "prior to public release" — so one agency's disclosure │ │ decision is subject to another agency's sign-off, by policy rather │ │ than by statute │ └─◆ UNDER A REGULATION THAT NAMES THE METHOD — and not a statistical agency │ health information held by covered entities, 45 C.F.R. § 164.514 │ ☐ ★ THE ONLY CORNER OF THE SYSTEM WHERE A MEMBER OF THE PUBLIC CAN READ │ THE RULE AND CHECK A RELEASE AGAINST IT. It governs hospitals and │ insurers rather than the agencies that publish the nation's │ statistics, which is the inversion worth stopping on: the regime │ with the least statistical ambition has the most legible rule. │ ■ PART D — WHO DECIDES, BODY BY BODY │ eight of the ten bodies checked have a formal review board; two do not │ ├─ CENSUS BUREAU ...... Disclosure Review Board — twelve voting members plus a │ non-voting programme manager, drawn one or two from each │ directorate, under the stewardship committee in Part B. │ "The DRB implements DSEP policy and makes policy │ recommendations, but it does not make policy." A division │ chief may appeal to that committee, whose decision is │ final. ☐ The only agency with a published charter for its │ board, and the only one with a published charter for the │ committee above it. ├─ NCHS ............... a board "chaired by the NCHS Confidentiality Officer"; │ the Director decides release. ⊘ "Board meetings and │ minutes are confidential." ├─ BLS ................ a board chartered under the Executive Oversight Board; │ its minutes go to that committee only. ⊘ No published │ charter and no published disclosure policy found. ├─ NCES ............... a board inside the statistical standards programme; it │ "will review the disclosure risk analysis report and make │ a recommendation to the Commissioner," and "The │ Commissioner then rules on the release of the data file." ├─ NASS ............... Data Access and Disclosure Review Board; review of the │ agency's own special tabulations is automated, enclave │ output is reviewed by hand under stricter rules. ├─ ERS ................ Data Disclosure Review Board — which, on its own │ account, "does not conduct disclosure reviews itself." │ It writes policy and trains the people who do. ├─ SOCIAL SECURITY .... a board inside the research and statistics office, whose │ disclosure-avoidance rules were approved by the CENSUS │ board as consistent with Census policy. ◈ One agency's │ internal standard certifying another's. ├─ BEA ................ ⊘ ◈ "No Disclosure Review Board." The functions are │ "covered informally, with a disclosure review officer," │ and internal research is "mostly self-enforced by │ program areas." ├─ TAX STATISTICS ..... ⊘ no board named in the agency's own statistical │ reporting manual and none found elsewhere; requests │ route through disclosure and liaison offices instead. ├─ EIA ................ ⊘ names no board in its own published account of how it performs disclosure reviews. │ └─ ★ ⊘ NOT ONE OF THE TEN PUBLISHES ITS BOARD'S DETERMINATIONS. Census is the only one that ever promised to publish anything in aggregate — its board policy commits to "an annual DRB report" whose "final version will be made available on the Census Bureau's external website" — and no such report was found on that website. ◈ Every number in every federal statistical table has been cleared by a body that publishes neither its reasons nor a count of its decisions. │ ■ PART E — WHAT IS PUBLISHED AND WHAT IS WITHHELD the line is not methods against parameters — it is counting rules against concentration rules, and every agency falls on the same side of both AGENCY OR PROGRAMME PUBLISHED AS A NUMBER WITHHELD ---------------------- ---------------------------- ---------------------------- census, enclave output cell size at least 3 the values of p and k census, enclave output round to 4 significant figs the concentration thresholds census, enclave output under 15 printed as under 15 the internal methods manual education statistics at least 3 unweighted cases the perturbation magnitudes health stats, files geography 100,000 or more the suppression criteria health stats, enclave frequency under 5 recoded the suppression rule itself agriculture, data lab at least 5 observations the values of n, k and p agriculture, data lab 30 weighted for demographic which rule hit which cell labour, employment the method, by name the threshold behind it labour, restricted nothing at all all of it; criteria on ask economic analysis the rule, by name the threshold, by decision energy information nothing at all the value, survey by survey tax statistics nothing at all no board, no rule, nothing ---------------------- ---------------------------- ---------------------------- 2020 decennial census the budget, the split, the nothing per-query allocation, the metrics file and the code Read the last row against all the others. The formal method applied to the 2020 census is the only case in the federal statistical system of an agency publishing its disclosure parameter in full — and it is the method the 2026 order removed. The most transparent technique in the system was the first one to go, and it went by an instrument that was never published. ■ PART F — WHERE EACH METHOD LANDS ON THE LADDER OF AGGREGATION the ladder is the one in the census geography map: nation down to the block METHOD WHAT IT DOES TO THE LADDER WHERE IT BITES ----------------- ------------------------------- --------------------------- suppression leaves it intact, deletes cells the smallest published cell coarsening moves the figure one rung up every rung, invisibly swapping holds the totals, moves people composition at the bottom noise infusion holds it, perturbs in place the small count formal privacy spends one budget across all every rung, by an amount synthetic file replaces the file, not the rung the user, not the table restricted access publishes nothing at that rung who can afford to travel Two consequences the method debate usually skips. First, coarsening and suppression are not alternatives to a geographic decision — they ARE one, so "which method" and "at which rung" are the same question. Second, the bottom rung is where every method converges and where the least is published: the continuous survey publishes nothing for census blocks at all, so the decennial block file is the only thing the block level ever carries, and the choice of method there is a choice about redistricting data and nothing else. ■ PART G — THE 2026 REVERSAL, AND THREE THINGS IT LEFT UNRECONCILED │ ├─◆ THE RECORD, IN ORDER │ │ 17 Jun 2025 .. the Census Bureau publishes its 2030 disclosure-avoidance │ │ research programme. A working group had compared ten │ │ candidate approaches — suppression, rounding, swapping, │ │ synthetic data, noise injection among them — and │ │ "concluded that formally private noise injection best fit │ │ its principles" for protecting block-level population │ │ totals, with state totals released as enumerated. │ │ 4 Jun 2026 .. DAO 216-26 takes effect and prohibits noise infusion. │ │ 4 Aug 2026 .. the 2025 one-year continuous-survey estimates are held: │ │ "The release date ... is being determined." │ │ 17 Aug 2026 .. the Bureau states that "a third method of disclosure │ │ avoidance called noise infusion is no longer allowed," │ │ that it "includes statistical methods such as record │ │ swapping and the use of synthetic data," and that "we will │ │ rely solely on coarsening and suppression methods." The │ │ June 2025 page is marked no longer current. │ │ 4 Sep 2026 .. more than eight hundred organisations ask the Secretary to │ │ release the held estimates on schedule. │ │ 29 Oct 2026 .. the estimates are released. The Bureau's own prerelease │ │ material records how: "Removed swapping to comply with │ │ DAO 216-26." │ │ ahead ........ a 2030 demonstration product in mid-2027, public feedback │ │ into 2028 decisions, and a first fully compliant five-year │ │ survey file in 2029 │ ├─◆ ≠ ONE. TWO INSTRUMENTS THAT CANNOT BOTH BE OBEYED │ │ A codified OMB rule requires the parent agency to let the statistical │ │ unit "determine the methods for conducting statistical activities." A │ │ departmental order tells the statistical unit which family of methods it │ │ may not use. The order's own savings clause says it "shall not be │ │ interpreted to conflict with any constitutional, statutory, regulatory" │ │ provision — and then routes any conflict to a Census deputy director to │ │ resolve "on a case-by-case basis." ◈ The instrument that might be │ │ unlawful appoints the official who decides whether it is. │ ├─◆ ⊘ TWO. THE AGENCY'S OWN PAGES STILL SAY OTHERWISE │ │ the published enclave methods handbook, current as of 1 May 2026, still │ │ offers researchers a noisy-proportion option and never mentions the │ │ order · the economic-census disclosure page still describes noise │ │ infusion for County Business Patterns · the population-estimates │ │ confidentiality page, revised 1 June 2026, still gives data swapping as │ │ its worked example · and the Bureau's live 2030 disclosure-avoidance │ │ research page, as read in November 2026, does not mention the order, │ │ noise infusion, coarsening or suppression at all, and gives no dates │ │ ◈ ONE CARVE-OUT IS DELIBERATE RATHER THAN STALE: the employment-dynamics │ │ programme keeps using noise infusion until the state data-sharing │ │ agreements that require it are revised, which the Bureau says "may │ │ take several years." │ └─◆ ★ THREE. NO COURT HAS EVER RULED ON WHETHER A METHOD IS LAWFUL │ The only case brought — a three-judge court in the Middle District of │ Alabama, decided 29 June 2021 — never reached the method. The State was │ held not to be "a person aggrieved" under the census statute's review │ provision and was dismissed with prejudice; the individual plaintiffs │ showed "no injury-in-fact sufficient to confer Article III standing ... at │ this point," on claims the court called "riddled with contingencies and │ speculation that impede judicial review"; and the injunction failed on │ delay, because "Plaintiffs waited approximately forty-two months." A │ procedural count survived and the case was voluntarily dismissed on │ 9 September 2021. │ ◎ SO THE QUESTION HAS NO FORUM. It is not in the Code, it is not in a │ rule, the rule that would have governed it is six years late, no court │ has reached it, and the body that answers it in practice publishes │ neither its minutes nor its reasons. ``` ## Notes - **What this map is drawn from.** Every statutory and regulatory passage quoted above was read at the text: 13 U.S.C. §§ 8, 9 and 214; 44 U.S.C. §§ 3504, 3506, 3561, 3563, 3572 and 3582; 26 U.S.C. §§ 6103 and 6108; 20 U.S.C. § 9573; 42 U.S.C. § 242m; 7 U.S.C. § 2276; 45 C.F.R. § 164.514; and 5 C.F.R. part 1321 as published at 89 FR 82453 (11 October 2024). Agency practice is from each agency's own published policies, standards, handbooks and conference papers, read 6 November 2026. The claim that no provision of the United States Code uses the phrases "disclosure avoidance", "differential privacy", "noise infusion", "cell suppression" or "data swapping" rests on a full-text phrase search of the Code, with the two returned hits read and confirmed to be word-level false positives. - **One instrument on this page was read at one remove, and it matters which.** The Commerce order's own page on the department's website was unreachable on every attempt in November 2026, so the three sentences quoted in Part B come from the Census Bureau's own disclosure-board policy, which recites the order among its legal authorities, and are corroborated word for word by a state budget and management office's September 2026 notice to its own data users. Two independent recitations in agreement is the best available footing, and it is not the order itself. - **Whether the ban reaches formal differential privacy is an interpretation, not quoted text.** No source found quotes the order naming it. What is on the record is that the Census Bureau has read the ban broadly — to cover record swapping and the use of synthetic data as a final protective layer — and that the Bureau describes its own 2020 method in the past tense. Probabilistic rounding survives as a non-deterministic form of coarsening, excludes 2030 census products, and needs product-by-product approval from the stewardship committee. - **The sharpest thing anybody involved has said out loud** came from a Census Bureau session at the August 2026 statistical meetings: "Finding that optimal balance is inherently a policy decision. It is not a technical one." The speaker was Michael Hawes, whose Census title is Senior Survey Statistician for Scientific Communication and who was the Bureau's senior adviser for data access and privacy from 2019 to 2022. Everything in Part B is a way of saying the same thing structurally: the policy decision is real, and it is made where the fewest people can see it. - **The published 2020 parameters, for the record.** For the redistricting file the Bureau published a privacy-loss budget of 17.14 for the persons file and 2.47 for the housing-unit data, a total of 19.61, set by the [[Data Stewardship Executive Policy Committee]] in June 2021 and announced with a spreadsheet of per-query metrics. The later demographic and housing file was published in a different accounting unit, roughly 7.70 for the household universe and 4.96 for the persons universe, in a technical paper by the Bureau's own staff; the two figures are not directly comparable to the first and the paper gives no equivalent in the earlier unit. - **Six questions deliberately left open.** Whether the overdue OMB standards rule has been proposed since its February 2026 target — no notice was found, and the agenda entry still reads proposed-rule stage. Whether the current edition of the enclave methods handbook retains the earlier edition's flat statements that "the values of p and k are confidential" — the relevant section could not be read in full, and the quotation above is from the 2021 edition. Whether an energy survey's planned 2026 switch to multiplicative noise survived the order, given that the Census Bureau conducts that survey for the energy agency. What the agriculture agency's suppression rule is for its *own published estimates*, as distinct from its enclave review. How many secure research centres there are — the Bureau's own page gives three different counts on one screen, so this map says roughly three dozen. And whether Congress itself specified any method for health records: the delegating section of the 1996 statute was not read at the text, so Part B says only that the method appears in regulation and not in the Code, which is a statement about where the words are. - **The two bodies and the one case this map rests on most heavily now have pages of their own.** The committee that chooses the Census Bureau's method is the [[Data Stewardship Executive Policy Committee]]; the board that clears each product under its authority is the [[Census Bureau Disclosure Review Board]], the only such board in the system with a published charter; and the single lawsuit ever brought against a disclosure-avoidance method is [[Alabama v. United States Department of Commerce]], which was decided without reaching it. - **Sample, not census.** Ten agencies and statistical units were examined for Part D and thirteen for Part C. The federal statistical system has more members than that, and this page does not claim to have drawn all of them; it claims that the design families found here are the design families that exist. ## See also - [[Census Geography Tree]] — the ladder Part F runs down, and the companion finding: the census block is the unit every district and precinct is assembled from, and the one unit the continuous survey publishes nothing for. - [[Census Data Access Crosswalk]] — the six routes by which a reader reaches census data, with the protection gate standing on each one. - [[Differential Privacy]] — the one method in Part A this wiki covers in depth, and the only one whose parameter was ever published. - [[Disclosure Avoidance for Statistical Products]] — the 2026 Commerce order that occupies Tier 4 of Part B. - [[Federal Statistical System Tree]] — the agencies in Part C as a structure, with the appointment mechanism beside each. - [[Statistical Officials Crosswalk]] — the official every agency owes by law, and how little the law gives one. - [[How a Federal Survey Is Approved]] — the clearance gate on the way in, where this map covers the gate on the way out. - [[How a Statistical Agency Is Defunded or Discontinued]] — the other eight routes by which a federal number stops meaning what it meant, one of which is a departmental order about method. - [[Data Stewardship Executive Policy Committee]] · [[Census Bureau Disclosure Review Board]] — the two bodies in Part B's Tier 4 and Part D, which between them decide the method and clear the product. - [[Alabama v. United States Department of Commerce]] — the only case ever brought against a statistical disclosure method, and the reason Part G ends where it does. - [[Census Data Products Tree]] · [[Census Bureau Org Chart]] · [[Federal Statistical Research Data Centers]] · [[Confidential Information Protection and Statistical Efficiency Act]] · [[Paperwork Reduction Act]] · [[Statistical Policy Directive No. 1]] · [[Chief Statistician of the United States]]