US hazard-resistant building codes are law only where a state or local government adopts them — the return the 2019 mitigation study put at USD 11 for every USD 1 lands on owners, insurers and the federal Disaster Relief Fund

What is happening?

FEMA runs a program called Building Code Adoption Tracking, which follows hazard-resistant code adoption across approximately 22,000 jurisdictions in the United States. Its most recent published reading is that 21 percent of jurisdictions have adopted current natural-hazard-resistant building codes and that 38 percent of the population is covered. The reading is dated only as of Q3, with no year given by either the publisher or the republisher; the FEMA file naming for the same release series carries FY25, which would place it at the quarter ending 2025-06-30. The program assesses five hazards — flood, damaging wind, hurricane wind, tornado and seismic — against model codes and the design standards ASCE/SEI 7 and ASCE/SEI 24.

The bar being measured moves on its own. FEMA treats a community as hazard resistant if it has adopted either of the two most recent editions of the International Code Council model codes without weakening the flood, hurricane wind and seismic provisions. The Code Council republishes those model codes every three years. A jurisdiction that adopted a current code once and then does nothing leaves the current set without repealing anything, without a vote, and without notice.

The value at stake is documented and one-sided. The National Institute of Building Sciences found in its 2019 mitigation study that adopting the current model code returns USD 11 for every USD 1 invested, at a national cost of USD 1 billion a year against USD 13 billion a year in benefit, and that moving to the 2015 model codes added about 1 percent to construction cost relative to 1990 standards. FEMA found in November 2020 that of 18.1 million buildings constructed between 2000 and 2016, only about half were built to the model codes at all.

Whose problem is this?

The body that decides and the body that pays are not the same body, and the distance between them is the shape of this problem. A council votes on a cost that lands this year inside its own borders, in exchange for a benefit that lands later, elsewhere, and largely on somebody who was not at the meeting.

Where does this problem end?

The boundary matters here because the model code already exists and costs nothing to copy. What is missing is not a text and not knowledge but a vote in about 22,000 separate places, and then the same vote again every three years.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names one. Not a share of jurisdictions, not a share of population, not a date. FEMA publishes this measurement on an annual cycle against no goal line, which means the measurement can report movement but can never report failure. The implied floor is the one the model codes already describe, adoption of either of the two most recent Code Council editions with the flood, wind and seismic provisions left intact, and every opened source is silent on when or how far.

How big is it?

About 211 million people. FEMA reports that 38 percent of the US population lives in a jurisdiction with a current hazard-resistant code, which leaves 62 percent outside it. Applied to the Census Bureau projection of 341,145,670 US residents at the start of 2025, that is about 211.5 million, reported here as 211 million.

The upper bound comes from a second federal statement. A White House proclamation of 2023-04-28 placed nearly two-thirds of Americans in communities that had not adopted the latest model codes, which against the same population base is about 225 million. The band is therefore 211 million to 225 million, and its width is not sampling error. It is the distance between two federal statements that use different definitions of current and different reference dates, and no opened source reconciles them.

This is a count of people by the legal regime over the place they live, not a count of people inside unsafe buildings. Most of the 211 million occupy structures built long before any current code, including inside the covered 38 percent, because codes reach new construction and substantial repair rather than the standing stock.

The live decision is therefore a flow rather than a stock. Privately-owned housing completions ran at a seasonally adjusted annual rate of 1,392,000 units in June 2026. If completions were distributed like population, roughly 863,000 units a year are being finished where no current code applies. Persons per unit is not derivable from any source opened here, so this figure stays in units rather than importing an unsourced household size, and the proportionality assumption is weak in a known direction — construction concentrates in the South, which holds both the strongest state code regime and large low-adoption areas.

Under what conditions does it arise?

1. The cost is local, immediate and attributable; the benefit is distant, statistical and partly leaves the jurisdiction. Adoption adds about 1 percent to construction cost and obliges a building department to review and inspect against a text it may not have staff for. The return of USD 11 for every USD 1 arrives as damage that did not occur in a storm that has not occurred, to an owner who may not be the buyer, and to insurers and the federal Disaster Relief Fund. Nothing in the local ledger records the avoided damage, because it never happens and never generates an invoice.

2. The measured bar advances on a three-year clock. Because FEMA counts only the two most recent Code Council editions, a jurisdiction that adopts once and then leaves the code alone is reclassified out of the current set by the calendar. The reported national share can fall while nothing anywhere is repealed.

3. The reward is scored one level above the body that adopts. In the BRIC grant program, building code activity was worth up to 40 of the 100 technical evaluation points, but CRS records that as of 2023-06-01 the majority of states and territories did not qualify for the building code point allotment, and that a community holding the latest codes could still be treated as noncompliant because its state did not meet the criteria.

4. States can cap local ambition. Some states limit local amendments to the state code or set conditions on them, which CRS notes may deter a local government from adopting requirements more hazard-resistant than the state floor. A jurisdiction that wants a stronger code may be barred from having one, and is then scored down for the choice its state made.

5. Adoption is divisible. A jurisdiction that adopts the model building or residential code may exclude some or parts of the ASCE design standards, so a place can be recorded as having adopted the model code while having removed the design loads that make it hazard resistant. The tracking program separately follows amendments that weaken adopted codes, and publishes no count of them.

What has been tried?

Two directions ran together for five decades — publish a model text that any jurisdiction may adopt for free, and pay jurisdictions to adopt it. A third direction began in 2026 and runs against the second.

What was found?

Why is it still unsolved?

Incentive inversion — the body that pays for a stronger code is not the party that collects the return, and five decades of federal machinery exist precisely to manufacture a local reason where the natural one points the other way.

A general model code is not law anywhere until a local body votes to make it law. That vote carries a price with names attached to it. Construction cost rises about 1 percent, the building department needs review and inspection capacity it may not have, and the people who pay both attend the meeting. The return of USD 11 for every USD 1 arrives as damage that did not occur, to an owner who may not be the buyer, and to insurers and a federal fund. The existence of an entire apparatus built to supply that missing reason — the code provisions of the Disaster Recovery Reform Act, the grant scoring, the code plus-up, the flood insurance class ratings — is itself the evidence that the underlying incentive runs the wrong way.

The second part is that the target moves and the reward is aimed above the level that decides. FEMA counts a jurisdiction as current only while it holds one of the two most recent Code Council editions, and a new edition arrives every three years, so a jurisdiction that does nothing is reclassified without acting. The most likely explanation for 35 percent of localities in 2022 becoming 21 percent of jurisdictions in 2025 is that the window slid, and no opened source states it. Meanwhile the money meant to supply the missing reason was scored at the state level: as of 2023-06-01 the majority of states and territories did not qualify for the building code point allotment, so a city that had done everything right could lose the points anyway. Appetite is plainly not the constraint. Across four years applicants requested USD 15.665 billion against USD 4.795 billion available.

The third part is what keeps the pattern stable. The single lever that was moving was ended on 2025-04-04, defended in court as never having been ended, ruled unlawfully terminated on 2025-12-11 in a judgment that compels no award, and reissued on 2026-03-25 with the state and territory code money at USD 56 million for two fiscal years where FY2023 alone had carried USD 112 million. Only 21.5 percent of the dollars ever selected had been obligated as of 2026-01-29. In the same season the federal government began removing the step at which a code is checked: Executive Order 14377 and the SBA interim final rule preempt a permit or approval requirement that delays disaster-loan-financed rebuilding by more than 60 days, substitute a two-page builder self-certification, and bar stop-work orders and penalties for the preempted requirement. The substantive code remains law and the gate that applies it does not. That gate is exactly where the one quantified loss in the federal record was incurred, when a quarter of USD 16 billion in insured Hurricane Andrew losses was attributed not to a missing code but to a code that was never enforced.

What observation would mean it is solved?

Candidates — (a) the tracked share of jurisdictions holding a current hazard-resistant code rises toward the whole country and stays there across at least two Code Council cycles (b) post-disaster damage assessments stop finding code-attributable losses in recently permitted buildings (c) federal disaster obligations for building damage fall relative to exposure.

(a) alone counts paper against a moving line. Adoption is divisible, so a jurisdiction may adopt the model code and strip the design standards that carry the hazard resistance, and the tracking program follows weakening amendments whose count is published nowhere. The indicator also has no stable base: an unknown portion of the jurisdictions outside the 21 percent sit in a category FEMA maintains for places where it has no code information at all, which is a different thing from a place known to lack a current code.

(b) alone is not being produced. Damage assessments happen where a disaster happens, and the assessments that exist are episodic rather than systematic. The one case in the federal record that attributes loss cleanly attributes it to enforcement, not adoption, which means this observation would be measuring the axis the national indicator does not cover. It would also take a generation, because the buildings that would carry the evidence have to be built first.

(c) alone moves with the weather and with the housing cycle. A quiet decade lowers obligations and a severe one raises them, independently of any code. Obligations also depend on declaration policy and on cost-share rules that change with an administration, so the series can move by several billion dollars without a single building being different.

The three have to be read together, and none of them is being produced on a schedule today. Two of the three would need a measurement layer that does not exist.

What is it connected to?

Fills with researchfederal disaster relief eligibility and whether it is blind to whether a jurisdiction has adopted a code, housing affordability policy and the cost objections that drive weakening amendments, the property insurance market and its own code-linked pricing, energy code adoption tracked separately by the Department of Energy, and code adoption regimes in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• The headline share circulates without its denominator, and its publisher is unreachable. The figures of 21 percent and 38 percent appear with no jurisdiction count and no population base attached. The count of approximately 22,000 jurisdictions lives in one CRS footnote, which adds that they cover most but not all of the United States without saying how much is left out, and no opened source multiplies the two. The FEMA pages that publish the number returned HTTP 403 to every attempt in this round. An archived copy of the underlying dataset has no readme and no data dictionary by its own metadata, and carries a FEMA notice telling users to confirm with their state or local code authority before relying on it.
• The reading has no year. It was published on 2025-10-08 and dated as of Q3, and neither the publisher nor the republisher states which Q3. The FEMA file naming for the same series says FY25, which would make it the quarter ending 2025-06-30, but no opened source says so. A national statistic is in circulation with an unresolvable reference date.
• The size of the unknown bucket is published nowhere. FEMA maintains a category for jurisdictions where it holds no building code data, distinct from jurisdictions known to lack a current code. No opened source states how many jurisdictions sit in it, so the 79 percent that is not adopting is a mixture of confirmed non-adoption and absence of information in an unknown ratio.
• Adoption is tracked and enforcement is not, and the record admits it in passing. One CRS report carries a section heading promising to cover the federal role in tracking both adoption and enforcement, and the answer beneath it describes only adoption. Nothing opened here reports a national inspection rate, plan review rate, compliance rate or building department staffing level, while the single quantified loss attribution in the whole federal record is to enforcement failure rather than to non-adoption.
• There is no target anywhere. Not in the tracking fact sheets as republished, not in the descriptions of the FEMA Building Codes Strategy, not in coverage of the grant notice. No share of jurisdictions, no share of population, no date.
• The adoption figures are never reconciled, and the mechanical explanation appears in no source. 35 percent of localities in 2020 and 2022, 21 percent of jurisdictions in 2025, 27 percent of building stock at FY2023 Q2, nearly two-thirds of Americans uncovered in 2023 — five statements on five bases. No source states whether adoption fell, whether the definition tightened as the code cycle advanced, or both. That the share can fall automatically every three years is a direct consequence of the published FEMA definition and is written down nowhere.
• The savings figures are projections and have never been checked against outcomes. The ratio of 11 to 1 dates from 2019, the projection of more than USD 600 billion by 2060 and USD 3.2 billion annually by 2040 from November 2020. Six years on, no opened source compares any projection with realized losses. FEMA states its own scope limit, that the counted losses are physical damage to buildings and exclude economic, social, cultural and government losses from loss of function, so the published benefit is a lower bound whose shortfall nobody quantifies.
• Nobody counts people, and nobody counts the local cost either. No opened source attaches a death, injury, hospitalization or displacement figure to the adoption gap; the unit of account throughout is avoided dollars of property damage. And while the construction-cost side is estimated at about 1 percent, no opened source states what adopting and enforcing a current code costs a small building department in salaries, training or review capacity, which is the quantity that actually decides a small-jurisdiction vote. The only national picture of that workforce is a self-selected survey from 2014.

See the evidence

No primary FEMA document was read directly. The number that anchors this problem is quotable here only at second hand, because every FEMA path attempted returned HTTP 403, and that limitation is itself part of the finding — a national statistic whose publisher is unreachable by machine, drawn from a dataset whose own archived metadata carries no readme and tells users to confirm it elsewhere. What was read directly: two CRS reports and one CRS Insight, rendered from the congressional PDF service; the NIBS mitigation fact sheet; the Harvard law program analysis of the SBA rule; the archived dataset metadata; and the Census construction release and population bulletin. The Federal Register texts at 91 Fed. Reg. 3989 and 91 Fed. Reg. 3813 were not opened in this round, so the executive order and the SBA rule are reported at one remove. Where sources overlap they agree — the reissued grant figures appear identically in two independent reports, and the limits of federal authority appear in two CRS reports written four months apart. Where they disagree the disagreement is left standing rather than resolved: the litigation is described as brought by 20 states in one source and 23 in another, the dollars at stake as USD 4.665 billion for 1,942 selected applications in one and nearly USD 4.5 billion for approximately 2,000 projects in the other, and the federal record itself contains three positions on whether the grant program was ever terminated — a FEMA press release ending it, a sworn FEMA declaration denying that it had ended, and a court ruling that the termination was unlawful. The adoption share is likewise left unreconciled at 35, 27 and 21 percent on three different bases. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 211,000,000–225,000,000 As of 2023 through a 2025 quarter

Sensitivity The width of the band is not sampling error. It is the distance between two federal statements about how many Americans stand outside current building codes, which differ by about four percentage points because one measures adoption of either of the two most recent model code editions without weakened hazard provisions and the other measures adoption of the latest model codes, at reference dates two years apart. What this number fails to count is the condition of buildings. It counts people by the legal regime over the place they live, not by the structure they occupy, and most of the 211 million live in buildings put up long before any current code, including inside the covered 38 percent, because codes bind new construction and substantial repair rather than the standing stock. It also counts adoption rather than enforcement, so a jurisdiction that adopted a current code and does not inspect is recorded on the covered side; given that no national measure of enforcement exists, this is likely the largest single distortion and its size is unknown. An unknown further portion of the uncovered share is not confirmed non-adoption at all but absence of information, because FEMA maintains a separate category for jurisdictions whose code status it does not know and publishes no count for it. In the opposite direction the figure overstates the live decision at stake in any one year, since the flow rather than the stock is what code adoption governs: privately-owned housing completions ran at a seasonally adjusted annual rate of 1,392,000 units in June 2026, of which roughly 863,000 would fall in uncovered jurisdictions if completions were distributed like population, an assumption that is weak because construction concentrates in the South, which contains both the strongest state code regime and large low-adoption areas. Persons per unit is not derivable from any source opened in this round, so that flow figure is deliberately left in units. Finally, the band does not intersect with hazard exposure. Roughly one third of the US housing stock is reported to sit at high risk of a natural disaster, and the overlap between high hazard risk and non-adopting jurisdictions is published nowhere and is certainly not independent.

Regional breakdown FEMA publishes the tracking result as national percentages and as state and territory resistance classes, not as population counts by region, and the FEMA pages and regional fact sheets that might carry per-region jurisdiction counts returned HTTP 403 to every retrieval attempt in this round. The one regional-flavoured table available, the FY2023 Q2 share of building stock by hazard, uses building stock rather than population as its base and its own footnote states that the flood row is drawn only from high-flood-risk jurisdictions, so its rows do not share a denominator with each other or with the population measure. Splitting the national figure by state population share would be apportionment, which invents a distribution the sources do not support, and is not done here.