# How the Census Is Actually Counted [[How the Census Apportions Congress]] begins at the moment a state population total exists and runs it forward into seats. This map is the other half — everything that has to happen before that number exists at all. It covers the stretch from *an address on a list* to *a state total ready for tabulation*, and it is the half where the political damage is actually done, because by the time the formula runs there is nothing left to argue about. The organizing fact is the one nobody expects: **the decennial census is an address operation before it is a people operation.** The Bureau does not go looking for people. It builds a list of every place a person could live, mails or delivers an invitation to each one, and then chases the addresses that did not answer. A person who lives at an address that is not on the list is not undercounted by a failed interview — no interview was ever attempted, because no one knew the housing unit was there. That is why the single most consequential stage in this entire pipeline happens in **2027**, three years before anyone fills out a form, and why the post-enumeration survey at the end can only measure the damage rather than repair it. The cluster this belongs to is routed from [[Census]]. Time-slice: operations and 2030-cycle schedule as of **September 21, 2026**, with the 2026 Census Test shown as it actually ran — it concluded August 31, 2026. > [!note] How to read this chart > `──▶` = the main path · `└─▶` = a branch, workaround, or attack surface · `◇` = a decision point · `★` = an operation that cannot be redone later · `✕` = a route that is closed or was dropped · `⚠` = a known quality problem with a number attached. Everything inside the fence is plain text; wikilinks live in the prose, Notes, and See also. Bracketed dates are the actual 2030-cycle schedule where one has been published, and `[live]` marks something in motion as of the build date. No officeholders are named — the pipeline does not change with who holds the seats, though as the Notes record, in 2026 it demonstrably changed with who held one. ``` HOW THE CENSUS IS ACTUALLY COUNTED — ADDRESS LIST TO STATE TOTAL │ Legend: ──▶ main path └─▶ branch ◇ decision ★ unrepeatable ✕ closed ⚠ quality problem │ Time-slice: operations as of Sept 21, 2026 · dates shown are the published 2030 cycle │ ■ 0. THE FRAME — YOU CANNOT COUNT A PERSON YOU CANNOT ADDRESS │ The count is built on the MASTER ADDRESS FILE (MAF), paired with the TIGER geographic │ database. Every later operation is a transaction against that list: mail to it, knock on │ it, impute for it. Nothing in the pipeline discovers a household that the list omits. │ ├─ ★ THE ASYMMETRY — a bad address is invisible; a bad interview is measurable. │ │ The post-enumeration survey at step 9 can size the error. It cannot add the people. │ ├─ 13 U.S.C. § 141(a) — the Secretary of Commerce "shall" take the census "in such │ │ form and content as he may determine." The manner of counting is DELEGATED │ │ DISCRETION, not constitutional command. Almost nothing below is in the Constitution. │ └─ ◆ WHAT THIS MAP DOES NOT COVER — what goes ON the form (a separate track that runs │ through OMB and Congress) and what happens to the number AFTER tabulation. │ ├─◆ 1. THE MAF IS MAINTAINED CONTINUOUSLY — all decade, not in census year │ ├─ USPS Delivery Sequence File ....... twice-yearly refresh of every deliverable address │ ├─ tribal, state and local files ..... via the Geographic Support System │ ├─ new-construction and building-permit data │ └─ ⚠ THE STRUCTURAL GAP — the DSF is a list of places mail is DELIVERED. Housing that │ receives no mail is systematically weak on it: informal and converted units, │ basement and garage apartments, units behind a single mailbox, colonias, │ and much of Indian Country. The gap is not random, and it is geographic, │ which means it lands as a seat and a funding formula, not as noise. │ ├─◆ 2. ★ LUCA — THE ONE STAGE OUTSIDE GOVERNMENTS CAN TOUCH THE LIST │ │ Local Update of Census Addresses. Tribal, state and local governments review the │ │ Census Address List for their own jurisdiction and submit additions, deletions, │ │ corrections, and spatial fixes. This is the ONLY point at which anyone outside the │ │ Bureau can put a housing unit onto the list on purpose. │ ├─ 2030 schedule as published: │ │ ├─ LUCA Prep — learn, gather materials, register ........ May 2027 – Aug 2027 │ │ ├─ ★ LUCA Review .................................. Oct 2027 – Mar 31, 2028 │ │ │ review window lengthened from 4 months (2020) to 6 months (2030) │ │ └─ all web-based for 2030; no paper materials │ ├─ THE PRICE OF ENTRY — a government must sign a confidentiality agreement and handle │ │ the address list under Title 13 to participate at all. │ └─▶ ◇ THE PARTICIPATION TRAP — LUCA is VOLUNTARY and UNFUNDED. A jurisdiction with a │ GIS department and staff to spare fixes its list. A poor, rural, or small one does │ not participate, or participates thinly. So the stage designed to correct the │ address gap is available in proportion to the capacity a place already has — │ which runs in the SAME direction as the gap it is meant to fix. Non-participation │ is invisible in the final data: there is no flag on a county that skipped it. │ ├─◆ 3. ADDRESS CANVASSING — verify the list against the ground [2029] │ ├─ IN-OFFICE (IOAC) ──▶ satellite and aerial imagery compared over time to detect │ │ housing change; the large majority of blocks are resolved without a visit │ ├─ IN-FIELD (IFAC) ──▶ listers walk only the areas in-office methods could not resolve │ │ └─ 2030 direction of travel: TARGETED canvassing, continuing the 2020 approach │ │ and downsizing the field operation further via automated change detection │ └─▶ ◇ THE TRADE — imagery sees roofs. It does not see the second family in the │ basement, the unit behind the unit, or a mobile home under tree cover. Every │ dollar saved by resolving a block in-office is a bet that the imagery was right. │ ├─◆ 4. ★ CENSUS DAY — ONE INSTANT, FOR EVERYONE [Mon Apr 1, 2030] │ │ Every answer describes the household AS IT EXISTED on this date, whenever it is │ │ actually given. An enumerator knocking in August 2030 is asking about April. │ ├─ ★ USUAL RESIDENCE — where a person lives and sleeps most of the time. Not │ │ citizenship, not voter eligibility, not legal domicile, not where the mail goes. │ ├─ the residence rule allocates the hard cases: prisons at the prison · college │ │ dorms at the school · nursing homes, shelters and barracks at the facility · │ │ children in joint custody where they sleep most · snowbirds at the usual home │ └─ ◆ WHY IT IS THE HIGHEST-LEVERAGE RULE IN THE SYSTEM — it is issued by the Bureau │ through notice-and-comment, not by statute. It moves whole populations between │ states and between districts, and it can be rewritten without Congress. │ ├─◆ 5. SELF-RESPONSE — cheap, fast, and the only stage the public controls [Mar–Jul 2030] │ ├─ INTERNET-FIRST since 2020 ──▶ phone ──▶ paper questionnaire │ ├─ delivery mode is set by address type: │ │ ├─ MAILOUT/MAILBACK ..... city-style addresses; the large majority │ │ ├─ UPDATE LEAVE ......... enumerator verifies the address and leaves a package │ │ │ (rural route and PO-box-only areas the mail cannot reach) │ │ └─ UPDATE ENUMERATE ..... remote areas and much of Indian Country; the lister │ │ updates the address AND interviews on the same visit │ ├─ up to five mailings, escalating, before an address is handed to the field │ └─ ★ EVERY SELF-RESPONSE IS AN ADDRESS REMOVED FROM THE MOST EXPENSIVE AND LEAST │ ACCURATE OPERATION IN THE CENSUS. This is why the whole communications campaign │ exists, and why self-response rate — not final coverage — is the number the │ Bureau manages against in real time. │ ├─◆ 6. NONRESPONSE FOLLOWUP (NRFU) — the expensive half [May–Aug 2030] │ │ Every address that did not self-respond goes to a temporary enumerator. Roughly a │ │ third of housing units nationally; far more in the places already hardest to count. │ ├─ the cascade, in order of quality: │ │ ├─ 1. household respondent — a resident answers .................... best │ │ ├─ 2. ⚠ PROXY — a neighbor, landlord, or building manager answers ABOUT the │ │ │ household. In 2020 only 87% of NRFU proxy enumerations were correct, │ │ │ and the item nonresponse was severe: 61% missing age or date of birth, │ │ │ 41% missing race, 38% missing ethnicity. │ │ └─ 3. administrative records — see step 7 │ ├─ contact attempts are capped; an address is not pursued indefinitely │ └─▶ ◇ THE COST LEVER THAT IS ALSO A QUALITY LEVER — cutting the number of visits, or │ shortening the NRFU window, converts household interviews into proxies and │ proxies into imputations. The count still finishes on time. The error moves │ downward into the small-area data and lands on the same populations every time. │ ├─◆ 7. THE PEOPLE THE ADDRESS MODEL DOES NOT FIT │ ├─ GROUP QUARTERS ENUMERATION — prisons, dorms, nursing homes, military barracks, │ │ group homes. Negotiated with the facility administrator, not the resident; │ │ often delivered as an administrative roster rather than individual interviews. │ ├─ SERVICE-BASED ENUMERATION — people with no usual address, counted at shelters, │ │ soup kitchens, mobile food vans, and outdoor locations over a set few days. │ ├─ TRANSITORY LOCATIONS — RV parks, marinas, campgrounds, motels. │ └─ ⚠ ALL OF THESE ARE ONE-SHOT OPERATIONS ON FIXED DATES. There is no second pass and │ no self-response fallback. If the facility roster is wrong or the outdoor count │ misses a site, the correction opportunity does not exist. │ ├─◆ 8. WHAT IS STILL BLANK WHEN THE FIELD CLOSES │ ├─ ADMINISTRATIVE RECORDS ──▶ IRS, Medicare, SSA, SNAP and other federal and │ │ commercial files used to resolve an address the field could not │ ├─ ★ COUNT IMPUTATION ──▶ for addresses where occupancy status or household SIZE is │ │ still unknown, the Bureau imputes from similar nearby units (hot-deck). In 2010, │ │ after an extensive field operation, this covered well under half a percent of │ │ addresses. The rate rises whenever the field operation is cut. │ ├─ ✓ IMPUTATION IS NOT SAMPLING. Utah v. Evans (2002) upheld hot-deck imputation for │ │ apportionment counts; sampling for apportionment remains barred by 13 U.S.C. │ │ § 195 and the 1999 decision. The line is narrow and it is load-bearing. │ └─ ◆ THE POINT — the census NEVER RETURNS A BLANK. Every address on the list produces │ a number. What varies is how that number was produced, and the further down this │ cascade it came from, the less anyone can see that from the published table. │ ├─◆ 9. PROCESSING → DELIVERY │ ├─ unduplication across the whole file (people counted twice, most often college │ │ students, children in two households, and second-home owners) │ ├─ DISCLOSURE AVOIDANCE — statistical noise injected into PUBLISHED tables under the │ │ Title 13 confidentiality obligation. State totals are published as enumerated. │ ├─ ──▶ tabulation of state totals ──▶ hands off to the apportionment pipeline │ └─ ★ POST-ENUMERATION SURVEY + DEMOGRAPHIC ANALYSIS — two independent measurements of │ how badly the count missed, run AFTER it is over. │ └─ ✕ NEITHER ONE EVER CHANGES A PUBLISHED NUMBER. They are a report card. │ Everything this map describes is final before the grade arrives. │ ╞═════════ [live] THE 2026 CENSUS TEST — WHAT ACTUALLY GOT TESTED ═══════════════════════════ │ First of two field tests before the 2028 dress rehearsal (test Census Day Apr 1, 2028). │ Ran May 1 – Aug 31, 2026; results not yet out as of Sept. 21, 2026. │ │ PLANNED (announced Jul 2024) AS RUN (revised plan, Feb–Mar 2026) │ ──────────────────────────────── ────────────────────────────────────────── │ 6 sites, incl. tribal lands in AZ, ──▶ ✕ 2 sites: Huntsville, AL · Spartanburg, SC │ western NC, western TX, (~155,000 housing unit addresses) │ Colorado Springs, CO │ nationally representative sample ──▶ ✕ dropped │ 2020 decennial questionnaire ──▶ ✕ the ACS questionnaire instead — ~40 min, │ incl. citizenship and education │ internet · phone · mail ──▶ ✕ internet self-response only │ bilingual English/Spanish paper ──▶ ✕ English only │ group quarters enumeration test ──▶ ✕ dropped, out of scope │ temporary Census field enumerators ──▶ ◆ USPS WORKERS used for in-field enumeration │ │ ◆ THE ONE GENUINELY NEW THING — letter carriers doing NRFU. The Bureau considered this │ for the 2018 end-to-end test and did not proceed, citing conflicting administrative │ processes between the two agencies. It is now the headline feature of the test. │ ⚠ EVERY STAGE THAT WAS CUT WAS A STAGE AIMED AT A HARD-TO-COUNT POPULATION: the tribal │ and remote sites, the Spanish-language path, the non-internet response modes, and │ group quarters. The design that remains tests the parts of the count that were never │ the problem. Congressional oversight interest is recorded in CRS IN12676 (Apr 2026). └──────────────────────────────────────────────────────────────────────────────────────────── ``` ## Notes - **The decisive stage is three years early and outside the Bureau.** LUCA review closes **March 31, 2028** — two years before Census Day. After that date the address list is effectively fixed, and every subsequent operation, including the entire field workforce, is a transaction against a list nobody outside the Bureau can add to. A jurisdiction that misses LUCA has missed the census, and will not find out until the post-enumeration survey three years later. - **LUCA's 2030 cycle is a reinstatement, not a routine renewal.** The operation went through a 60-day *Federal Register* comment period in early 2026 with comments due April 20, 2026. The review window was lengthened from four months to six in response to requests for more time. - **The 2026 Census Test was cut from six sites to two.** The initial design was announced in July 2024; the revised design was released in February 2026 and elaborated in March 2026. The Bureau's own statement describes the test as focusing "on operations in two of the planned test site locations." The dropped sites were the tribal-lands, remote-rural, and additional metro sites selected specifically for their hard-to-count populations. Precedent exists — several 2017 test sites were cancelled during the 2020 cycle over funding uncertainty — but the 2026 cuts were not attributed to funding. - **Using the ACS instrument in a decennial test puts a citizenship question inside a decennial test.** The [[American Community Survey]] has asked citizenship since 2005 and asking it there has never been the contested question; asking it of every household on the decennial form is. The 2026 test substituted the ACS questionnaire for the 2020 decennial questionnaire, so the test's self-response instrument runs roughly forty minutes and includes citizenship. What that does or does not imply for the 2030 form is not established by the test itself — the 2030 form is set on the separate § 141(f) reporting track, whose deadlines are April 2027 for subjects and April 2028 for questions. See [[Census Citizenship Question]]. - **Proxy response is the quality cliff, and it is the first thing a schedule cut produces.** The 2020 figures are the ones to keep: 87% of NRFU proxy enumerations correct, 61% missing age or date of birth, 41% missing race, 38% missing ethnicity. Proxies are not a small residual — they are the standard outcome when an enumerator cannot reach a household within the contact cap. - **The census never returns a blank, which is what makes the quality cascade invisible.** A published table does not distinguish a household interview from a proxy from an administrative-record fill from an imputed count. All four arrive as the same integer. See [[Census Undercount]] for what the difference costs and to whom. - **Disclosure avoidance is a separate fight with a separate clock** and is not resolved by anything on this map. See [[Differential Privacy]]. - **Verification note:** all 2026-test and 2030-LUCA facts here were verified against Census Bureau releases and a Congressional Research Service Insight dated April 1, 2026, as of September 21, 2026. No current officeholders are named in the chart. The 2026 test's *results* had not been published as of that date. ## See also - [[How a Census Question Is Added or Removed]] — the other track this map's block 0 sets aside: what goes ON the form, and who decides it - [[Census Data Products Tree]] — the other downstream branch: what the tabulated totals are actually published as, and which of those products any statute obliges the Bureau to deliver. - [[Census]] — the routing hub for this cluster - [[How the Census Apportions Congress]] — the downstream half: state totals through certified seats - [[Census Bureau]] · [[Department of Commerce]] · [[Federal Statistical System Tree]] - [[Census Undercount]] · [[Census Citizenship Question]] · [[Differential Privacy]] · [[American Community Survey]] - [[2020 Census]] · [[2030 Census]] · [[2030 Census Residence Rule]] - [[How a Redistricting Map Is Drawn and Challenged]] · [[Congressional Apportionment Mathematics]] - [[Foundations Hub]]