This deck is built for a screen
At this width a 16:9 slide shrinks below anything readable. The same fifty-four beats are set as a page, and that version is built for a phone.
Rotate to landscape, or open this on a laptop, for the deck itself.
Housing as a Product · Doc HAP-08 Rev 1.0 · August 2026
It takes about thirty-two months to deliver an American home. Eight of them are spent building it.
Working thesis · No gate cleared · Every claim graded and sourced · Rev. Aug 2026 · John Mullin · the research site
Movement 1 of 5
The world as it is, and the fact that breaks it
It takes about thirty-two months to deliver an American home.
Each square is one month on the critical path, from site identification to closing: feasibility, entitlement, design and review, site work, vertical build, sale.
InvestorsThirty-two months is longer than the learning cycle any risk-capital pool is built to tolerate. Hold that number — it is the one that decides who is allowed to fund housing.
PolicymakersNo dataset in the country tracks this figure end to end. It is assembled from separate sources that were never designed to be added together.
ResearchThe figure is assembled, not measured. Register row E-038 records that absence as a finding rather than a footnote: Census measurement begins at authorization, and nothing spans site identification to closing.
Which of those months are actually building the house?
Eight of those months are spent building the house. The other twenty-four are paper.
Eighteen months pass before physical work begins. Four are site work. Eight are the house going up. Two are the sale. The black squares are not the problem — they are the product. Everything else is overhead.
InvestorsEvery prior attempt to industrialize housing went after the eight. The twenty-four have never been touched.
PolicymakersThe part that is slow is the part government administers. That is not an accusation — it is an opportunity, because information processing is the one input that has become dramatically cheaper.
ResearchCensus publishes no single authorization-to-completion figure. Two averages for 2025 add to 8.8 months — 1.4 to start and 7.4 building, both for all 1-unit buildings (E-002). The pre-permit phase dominates the timeline and is not physical work. What this beat said before C-016 was wrong three ways, and the 2022 figures that replaced it went stale when Census resumed publishing; the corrections log has both.
Three quarters of the clock, and none of it building. That much waiting is not free.
Somebody carries those twenty-four months, and the line they are carried on is thin and getting thinner.
Somebody pays for those twenty-four months, and the builder is the first place to look. The second-largest homebuilder in the country reports gross margin on a house falling every year for four years — 29.5% in mid-2022, 22.6% in 2024, 17.8% in 2025, 15.6% in the quarter ending May 2026 — while the incentives it hands back to move those houses rose to 14.3% of revenue. Across seventeen years of its own cost surveys the industry's trade association puts the builder and developer gross profit rate at 9.6%. Whatever the clock costs, it is not coming out of a line like that. It comes out of the projects that never start — an inference from the margin, since nothing here counts them.
InvestorsThe incentive is the tell. A firm that could hold price does not hand back a seventh of revenue to move product, and it buys the mortgage rate down rather than cutting the sticker because the buydown costs about a quarter as much for the same effect on the buyer. That is a business absorbing a constraint, not setting one — and it means margin compression is where the cost is showing up, not where it is coming from.
PolicymakersThis is the beat that rules out the easy answer. If new homes were unaffordable because builders were taking too much, margins would be widening. They are compressing, and they are compressing in numbers filed under securities law by the firms doing it. What builders themselves name as their serious problems is the cost and availability of developed lots, and of labour — both upstream of anything they control.
ResearchRead the limits before the conclusion. This is one builder over fourteen quarters (E-222) set beside a trade association's own long-run assumption (E-224), and that association advocates deregulation, so a thin builder margin is a figure that helps its case — which is why E-224 is graded Supported rather than Confirmed. The two are also not the same measure: a public company's gross margin on home sales and NAHB's gross profit rate for builders and developers are differently scoped, and neither is a return on capital. So this beat claims a negative and only a negative — the cost of the clock does not fit in the builder's margin. Where it does sit is not established here. And the headline's forward claim, that a compressing margin prices production out, is an inference from the margin rather than a measurement: nothing on this page counts the projects that never began.
A margin that thin cannot absorb much. And one cost line is already larger than all of it.
Regulation accounts for roughly a quarter of the price of a new American home.
$131,734 of a $499,500 average new-home price — larger than the margin the builder keeps. Lennar’s gross margin on home sales, read from its own 10-Q, was 15.6% in the quarter ended 31 May 2026 (E-222). The headline compares a trade association’s survey against one listed builder’s filing, and both should be read as what they are. Meanwhile home equity is the median American household's dominant asset and its retirement plan, and roughly two thirds of households own. That single fact sets the politics of everything downstream.
InvestorsA cost line that large, sitting in the one part of the process nobody has automated, is the definition of an addressable market.
PolicymakersHousing is now the top issue for voters aged 18 to 34 going into the midterms. The constituency for this exists; the instrumentation to serve it does not.
ResearchThe 26.4% figure comes from two self-report surveys of the trade association's own members — 54 of 2,125 land developers responding, plus 337 builders — and the association advocates deregulation, so it is graded Supported rather than Confirmed. It is carried here because it is the best published estimate, not because it is disinterested.
A quarter of the price is paperwork. On a thing of what size, though?
A sixth of GDP. $48 trillion of standing assets. And a shortfall measured in millions of homes.
A fourth figure used to sit here: Hsieh & Moretti’s estimate of US growth forgone to housing supply constraints. It has been removed. A peer-reviewed Comment — Greaney, AEJ: Macroeconomics 18(2), April 2026 — replicated the paper, documented errors in its code, found the counterfactual as coded lowers output, and put the corrected effect roughly two orders of magnitude below the published one. The claim is register row E-054, graded Contested, and it is no longer quoted on this site as a number. The dispute is the finding.
Housing is about 16% of US GDP (E-052) — 12.3% of it the value of living in homes and 3.7% the work of building them. That 3.7% is what this argument is about, and it is the part that has been shrinking. The dollar conversions this beat used to carry rested on a national-accounts total with no register row, and are gone rather than re-derived. Standing behind it is $48 trillion of owner-occupied housing: the largest asset class in the country, and where most owning households hold their dominant asset. Against all of that, the country is short of demand by millions of homes, and the best-known estimate of what supply constraints cost in forgone output runs to double-digit percentages of growth over four decades.
InvestorsThe addressable surface is not housing's whole share of GDP. It is residential fixed investment — 3.7% of it — plus the transaction and financing layers on top, still one of the largest under-automated processes in the economy.
PolicymakersThe shortage figure is the one to hold, and to hold loosely: estimates range by method and definition, and the Congressional Research Service has reviewed why they disagree. What is not disputed is the direction or the order of magnitude — millions of homes, sustained over years.
ResearchThree grades on one slide. The GDP share is Confirmed but published by a trade association from BEA data, and Q1 2026 already reads 15.9%. The dollar figures are rounded on purpose: they are arithmetic on the shares, and BEA publishes growth rates rather than levels, so the denominator wants verifying before anyone quotes a precise number. The shortage is Supported and belongs in a range, never as one number. The output cost is Contested, and the correction printed under the chart is why.
A sixth of the economy, and short by millions of homes. How many, exactly?
A standing stock of 147 million against a need of 150.7 million. The difference is homes that do not exist.
Freddie Mac's Q3 2024 decomposition, and the eleven-study spread reported in the same capture (E-053). Both named primary sources refused automated retrieval, so the row is snapshotted from a secondary and graded Supported. The register's instruction on it is to cite the range, never the point.
Freddie Mac puts the standing stock at 147.0 million housing units against 150.7 million needed — a shortfall of about 3.7 million homes. Roughly 1.0 million of that is households that never formed: adults living with relatives or sharing, because nothing they could afford came onto the market. The other 2.7 million is the vacancy slack a working market needs and does not have. Up for Growth reaches 3.78 million by a different method. Eleven studies of the same question span 1.0 million to 4.4 million. They disagree about the size of the deficit. None of them finds a surplus.
InvestorsThe demand is already counted and already unmet. Nothing in this argument depends on creating it — only on whether supply can respond to it, and what follows is more than half a century of evidence that it cannot.
PolicymakersThe number under the number is the one to carry: about a million households that do not exist because the homes do not. They are not on a waiting list and they do not appear in a vacancy statistic — they are living with relatives, and they are the constituency for anything that raises production.
ResearchCite this as a range: eleven studies return 1.0 to 4.4 million, and the instruction on E-053 is never to quote one settled number. Read the decomposition too — 2.7 of Freddie Mac's 3.7 million is vacancy adequacy, not families with nowhere to live. Both named primaries refused retrieval; the row rests on a secondary capture.
Something that far behind has every reason to have got better.
For thirty years American housing industrialized faster than cars. Then it stopped, and it has never recovered.
Shape as reported in NBER 33188: stagnant 1900–40, a boom to 1970 in which homes per worker at times outgrew cars per worker, then a collapse. Schematic — the curve shows the finding, not plotted values.
Homes per construction worker were flat from 1900 to 1940, rose steeply to 1970 — at times outgrowing cars per worker — and then fell. Manufacturing kept going and never looked back. The 2008 shock made the divergence concrete: motor vehicle assemblies fell 47% and were back at their 2007 level four years after the trough, while housing completions fell 61% and took thirteen. Whatever changed, it changed around 1970, it changed only in this industry, and more than half a century of technology since has not undone it.
InvestorsEvery input has improved since 1970 — materials, machines, software, logistics, capital. Output per worker did not. That gap is where the value is, and it is why the constraint cannot be a technology gap.
PolicymakersThis is the fact that should trouble a legislator most, because it rules out the easy explanations. Not a labor shortage, not interest rates, not a bad cycle: a plateau that has held since 1970 and that no boom has lifted (E-015).
ResearchThe best causal work attributes the plateau to project-level regulation capping firm size — a competing diagnosis this document has not resolved. A separate dispute holds that part of the measured decline is a price-deflator artefact. Neither is settled, and both are on the record here. The recovery comparison is arithmetic on two published series and carries no causal claim: the vehicle industry received a federal rescue in 2009 that housing did not, and E-226 records that confound on the row rather than leaving it to the reader.
Everything else got better and housing did not. Then, for the first time anyone can point to, one number moved.
By June 2026 the median existing home sold for $446,400 and the median new one for $398,300.
The chart defaults to five years, where the premium collapses in order: new stood $26,400 above existing in 2021, $41,700 in 2022, $34,500 in 2023, $7,800 in 2024. Switch it to thirteen months and the inversion stops looking clean. Two of the three offered explanations are not improvements. Builders are buying the price down — 14.3% of revenue in incentives at the second-largest builder. And the house has shrunk, from a median near 2,700 square feet in the mid-2010s to about 2,400 today: a price that falls because the product got smaller is not a productivity gain. The third is not about new homes at all — owners on 4% mortgages will not sell into a 6.5% market.
InvestorsThe discount is bought, not earned. Fourteen points of revenue is margin, and a 300-square-foot contraction is a smaller unit at a similar price per foot. Neither is a cost curve bending. What it establishes is that builders can move price when they choose — that constraint was demand, and it was solved in eighteen months.
PolicymakersThe reversal is not evidence the supply side has been fixed. Two of the three mechanisms are builders absorbing or shrinking their way to a number; the third is a lock-in effect no housing policy chose. As an outcome measure it says almost nothing about whether building got easier.
ResearchThe decomposition has not been done. E-215 is Untested precisely here: nobody has separated how much of the gap is incentive, how much is square footage, how much is mix. The inversion is a fact without a mechanism, and E-214 rests on a series this project has not seen.
The first number to move moved the right way, and it still did not close the gap that matters.
The median American home now costs five times the median household income. Within living memory it was three and a half.
Endpoints are the register's sourced values — $82,800 against $23,620 in 1985, $416,900 against $83,150 in 2025 (E-072). The paths between are schematic: drawn from the shape of the underlying series, not plotted from them.
In 1985 the median home cost about three and a half times the median household income. It now costs about five. Half of America's renters — 22.7 million households, the most ever recorded — spend more than thirty percent of their income on housing, and the median first-time buyer is forty years old, up from the late twenties in the early 1980s, holding the smallest share of purchases ever recorded. The gap is simple, it is measured, and it is widening.
InvestorsDemand is not the question in this market and never has been. The question this deck answers is why supply cannot respond — and why the constraint is priced by nobody.
PolicymakersThese are the government's and the industry's own long series — Census, NAR, Harvard's Joint Center. They disagree about causes. None of them disputes the gap.
ResearchThree rows, three grades, stated up front. The ratio joins two primary series through secondary compilations and is Supported. The renter-burden count is Harvard JCHS and Confirmed. The buyer age and share are a trade association's own survey, methodology consistent since 1981, Supported. This beat states the symptom and claims no cause — what share of it supply constraints explain is Contested, and the register says so (E-054).
Not the builder’s margin, and not a shortage of technology. So where? It is built into how the work itself is organised.
Movement 2 of 5
Six parties, one set of facts, and not one of them kept
The industry is not a line. It is a mesh in which everyone rebuilds the same facts and nobody is authoritative.
An architect, a planner, a plan reviewer, a utility engineer, an appraiser and an insurer all need the same facts about the same parcel — dimensions, easements, utility access, flood status, setbacks, what is permitted. Every one of them reconstructs those facts independently, none of them is authoritative, and disagreement is resolved by meeting. Six parties is fifteen possible conversations. Give them one record and one determination to read, and it is six.
InvestorsThis is the pathology behind every other one. Six independent reconstructions of the same dataset is not inefficiency at the margin — it is the reason a project cannot be a product, and the reason the tenth costs what the first did.
PolicymakersNote what the right-hand picture does not change: the same six parties, with the same authority, asking the same questions. What changes is that they stop each producing a private answer. Singapore's eight agencies read one national schema, which is why they can answer together inside a twenty-working-day ceiling.
ResearchThe claim that this is the binding pathology is argued rather than measured. Layer 3's falsification test sits exactly here: if most measured delay turns out to be jurisdiction queue time rather than applicant resubmittal from inconsistent facts, a shared record delivers considerably less than this slide implies.
One record instead of six. Now: what does the tenth project cost?
The same journey, twice. Same standards, same reviewers — the difference is whether anything survives to the next one.
Today: land is optioned before anyone knows what is permitted, the design is drawn bespoke, review runs by hand and in series, appraisal needs comparables that do not exist, lending is priced to the worst case, building runs stop-start, and the sale is a bid against scarcity. Then the next project starts from zero. Rebuilt, the same eight steps run against a certified type and a shared record — and the loop returns to the type rather than to nothing.
InvestorsThat loop-back arrow is the whole thesis in one mark. An industry where the tenth unit costs what the first did has no products, only projects, and no compounding of any kind.
PolicymakersCount what is unchanged: the standards, the reviewers, and every check that gets made. What changes is that the first three steps stop being redone from scratch, one step — the determination — is new, and the loop at the end returns to the type instead of to nothing.
ResearchThe claim is about marginal cost, and it is not measured here. That the tenth project would be cheaper than the first follows from reuse rather than from evidence — no American jurisdiction has produced the data because no American jurisdiction has the arrangement.
Nothing survives to the next project because nothing is shared. So what, exactly, is fragmented?
How it is built has one answer. Where it may go does not.
The split is written in statute, twice over. Federal law preempts every state and local construction standard for a certified manufactured home — one code, fifty states (42 U.S.C. §5403(d)). Arizona's preemption section draws the same line, then names what stays local: zoning, fire zones, setbacks, maximum area and fire separation, site development, property lines (A.R.S. §41-4006(D)). The fork is not in how these homes are built but in where they may go — and how many siting regimes there are, and how far they diverge, has never been measured (E-131).
InvestorsThe codes beneath this fork drift too — adopted editions spanned 2012 to 2021 as of 2024, version pinning with no resolver (E-024, E-025), and both measured revision cycles ended larger (E-133). But siting is the load-bearing layer: it decides whether the certified product may exist on a parcel at all.
PolicymakersWhen a jurisdiction amends its siting rules, nobody can compute which parcels or in-flight projects are affected. The blast radius is unknown until it is felt.
ResearchTwo legs, different grades. That the split exists is Confirmed from statutory text — §5403(d) federally, §41-4006(D) in Arizona, both quoted verbatim in the register. Its magnitude — how many regimes, how far apart — is E-131, and it is empty. The beat asserts the first and only the first.
And nothing about the process even requires it to be slow.
The same six reviews, reading the same facts, at the same time, with a deadline.
Today each gate needs the previous gate's output as its input, so they execute in series. Any one can block indefinitely, none has a timeout, and nobody is accountable for the total. Given one authoritative record every gate can read, the same six reviews run concurrently and return a single consolidated response — each under a statutory deadline with a defined default if it is missed.
InvestorsThis is not a proposal. It has been running in production in Singapore since December 2023 across eight agencies — mandatory for large projects since last October, and for all new projects this October.
PolicymakersThe objection this diagram answers is the one that matters most: that speed comes out of scrutiny. It does not. The same reviews happen, by the same reviewers, against better facts. What is removed is the queue between them — and Indiana already has the template for the default remedy, giving applicants a private-review fallback when the jurisdiction misses its window.
ResearchLayer 4's falsification test lives here: if measured delay proves to be jurisdiction queue time rather than applicant resubmittal, this delivers far less than claimed — and if jurisdictions systematically miss the deadline and the default proves politically unenforceable, it produces litigation rather than throughput.
Reordered, not removed. But is a review you did not repeat still a review? Software answered that question decades ago.
Eight principles every mature software system has in the world of development online, we can learn from and apply to the world of housing or development offline to reduce these clear inefficiencies.
No static analysis, so every error surfaces in production. No type system, so invalid states are representable. No shared state, so six parties reconstruct the same parcel facts independently. Blocking sequential calls with no timeouts. No regression tests, so amendments break unknown things. No package registry. No demand signal. No observability.
InvestorsThe cost of a defect rises by roughly an order of magnitude at each stage it survives. Housing catches its defects at the most expensive stage available.
PolicymakersThe last item is the one that should trouble a legislator most: states pass reforms and then have no instrumentation to determine whether they worked. You cannot optimize what was never measured.
ResearchEvery capability on this list has been standard in software engineering for decades. None is exotic, and none has ever been applied to housing production. Singapore's shared national schema is the one production counter-example.
Eight principles, none of them adopted here. Every one has been built before — and one city built the whole thing.
It worked. It reached industry use within a few years. And it covers part of a code, in one jurisdiction.
FIXEDCompiled rules, shared schema, 7-agency review in 20 days
KILLED BYHand-encoded rules, part of one code — and four transplants that never shipped
Singapore built the first successful automated code-compliance system anywhere: CORENET in 1995 for 2D plans, upgraded in 2002 as e-PlanCheck to read 3D models. It was commissioned by the building authority and has been commercially used. What it checks is part of the code — architectural and building services — not a building code entire. Its current generation runs eight-agency concurrent review against a shared national schema, commits to a consolidated response within twenty working days as a service ceiling rather than a measurement, and becomes mandatory for all new projects in October 2026. Its rules were hand-encoded, and users cannot author new ones.
InvestorsThis site used to tell you the price of that: twenty-eight years for one jurisdiction, and a multiplication that closed the market. That was wrong, and what replaced it is at the end of this act.
PolicymakersThis is the existence proof that compiled codes work in production, with a regulator's safety record intact. It is mandatory law in a developed country this October. Read the scope honestly: a part of a code, not the whole one.
ResearchSeparate two things. The hand-encoding constraint generalizes and still holds (E-023). The claim that its cost stopped replication does not. And E-079 is load-bearing on one peer-reviewed source, which says the system "has been commercially used" without dating first industry use — Supported, corroboration outstanding.
So it works, once, in one place. America has done its own version of exactly this, and it was not software.
Movement 3 of 5
The one time this country made a house cheaper on purpose
In 1976 Congress federalised how one kind of house is built, and it produced the least expensive way to build a home in America.
The National Manufactured Housing Construction and Safety Standards Act of 1974 took effect in 1976, and it did the thing this whole argument says is hard: it wrote one construction standard and preempted every state and local standard on the same aspect of performance. One code, one determination, fifty states. What came out of it is the cheapest new housing the country has ever produced — the buyers are the households the rest of the market has priced out. Emily Hamilton told Congress in March 2025 that HUD-code housing is the least expensive way to build a new home in the United States, and that the median income of its buyer is $57,000 against $93,000 for site-built.
InvestorsThis is the closest thing to a controlled experiment American housing offers. Preemption was tried, on one product, and the cost came down. Whatever else is uncertain here, that half is not a hypothesis.
PolicymakersEverything a preemption sceptic predicts — unsafe homes, a race to the bottom, communities overruled — was the case against the HUD Code in 1974. What actually happened is that the homes got cheaper and the standard got enforced federally. The failure came later and from somewhere else entirely.
ResearchE-104 is Supported, not Confirmed, and the wording matters: Hamilton told Congress this. It is expert testimony in a congressional record, not a measurement, and her footnotes for the income comparison have not been chased. E-097's price figure is Census and Confirmed, but the 2025 annual number on that row is an unweighted mean of monthly averages and is labelled as such.
That is one half of a housing system, built and working. The boundary of what it settled is exactly where this story turns.
The 1976 Act preempted how a certified home is built — and stopped there. What that home legally is, and where it may go, never got a standard.
The Act's preemption clause covers construction and safety standards for the same aspect of performance — nothing more (E-091). Arizona shows the boundary inside one section of state law: construction and installation preempted, then zoning, setbacks, site development and utility terminals expressly left local (E-132). And on a third question — what the finished home is as property — the Act is silent altogether. So the precedent is real, and it is half-finished by its own text: one national answer for how, no standard for what or where. How much the missing half costs is not rhetorical; it is unmeasured (E-131).
InvestorsThe asymmetry is the opening: certify-once-and-travel is proven for construction and absent for placement. What that absence is worth is E-131, and it is empty — Gate 0 is the two-week measurement that starts filling it.
PolicymakersArizona is the live case and the uncomfortable one, and it is worse for this argument than this site used to say. Statewide by-right accessory-dwelling rights, automatic loss of local control on a missed deadline (E-027) — and since 2025 a statutory duty on every municipality to publish standard preapproved plans, accessory-dwelling sizes included, beginning 1 July 2026 (E-211). The right was granted. The plans were mandated. Mesa issued 29 accessory-dwelling permits in the first eight months of 2025 against 30 in all of 2024 (E-210). What is still missing is a shared parcel record, a published series, and any assurance that a home a buyer can order may be placed at all — a city can grant the right and exclude the cheapest house from it in the same ordinance, and one does (E-208) — on authority the state expressly reserved to it, with no Arizona statute located that would bar it (E-209).
ResearchThe realistic American first-best is not national zoning — land use is state police power delegated to localities. It is state-level preemption, Auckland's model rather than Tokyo's, and it is already moving in Montana, California, Washington and Arizona.
The property silence first. America built its cheapest housing — and let fifty states call it a vehicle.
Movement 4 of 5
What happened next to the cheapest housing America knows how to build
The Act settled how these homes are constructed and said nothing about what they are. States answered that question with a motor vehicle title.
Read the 1974 Act for what it does not say. It preempts construction and safety standards; it is silent on whether a manufactured home is real property or personal property. That silence was not neutral. States classify these homes through motor vehicle agencies and issue a certificate of title, and roughly three-quarters of them have since written statutes providing a procedure to convert one into real property — a procedure that only has work to do where the default is personal property. So the cheapest new house in America is, in most of the country, bought the way a car is bought.
InvestorsThis is where a good policy was quarantined. Not by opposition, and not by a later repeal, but by a classification nobody argued about, in a layer the reform did not touch.
PolicymakersThe justification for treating the home as a vehicle was the permanent chassis: it can be moved, so it is transportable property. The chassis was required by the federal definition until July 2026. What share of these homes is ever actually moved after placement is a figure this project could not source, and it is not claimed here.
ResearchTwo legs, and they do different work. E-091 is the statute, read directly: 42 U.S.C. 5403(d) preempts construction standards on the same aspect of performance and addresses property classification nowhere. That establishes federal silence, not state practice. E-120 is state practice, from the CFED and NCLC fifty-state review: "approximately three-quarters of the states have statutes that set forth a procedure to convert a manufactured home from personal to real property", with the states listed in its Appendix A. The count this site once carried — 42 states issuing certificates of title — is withdrawn, because the obvious authority does not state it (E-092). And the most quotable fact in this beat is one this project refuses to quote: see /literature on why.
Once the cheapest house in America is legally a vehicle, it gets financed like one.
Manufactured-home purchase applications are denied 42% of the time against 7% for site-built. The gap survives controlling for credit score.
Chattel paper is not a mortgage. It carries a higher rate, a shorter term, weaker consumer protections and no secondary market to speak of, and it is what the vehicle title produces. In 2019 HMDA data, 42 percent of manufactured-housing purchase applications were denied against 7 percent of site-built. Only 27 percent of manufactured-home applications resulted in a loan at all, against 74 percent. And the disparity is not explained by who is applying: in the super-prime band above 720, chattel approval runs 63 percent and MH-mortgage 80 percent against 95 percent for site-built, and landowners who take chattel loans have similar credit scores and incomes and better loan-to-value and debt-to-income ratios than landowners who take mortgages. The paper refinances almost never — under 4 percent of originations against 44 percent site-built (E-010) — at spreads the register carries as 5.61 against 1.20 percent (E-011).
InvestorsA product does not lose five sixths of its volume because buyers stopped wanting it. It loses that because the financing that would have carried the demand does not exist for it, and the reason it does not exist is a property classification.
PolicymakersThe CFPB's own reading is the load-bearing sentence: the evidence "suggests that chattel borrowers' credit profiles would not have prevented them from getting a mortgage". Whatever is happening at the point of refusal, it is not borrower quality. That moves the diagnosis from risk to product architecture, and it is the strongest causal claim on this site that this project did not have to construct.
ResearchThree measures, three rows, and they are not interchangeable — E-112 is the denial rate, E-113 is the credit-banded approval rate as the report defines it at footnote 32, and E-009 is the origination rate. This site carried all three inside one row under the word "approval" until August 2026, which is correction C-014. Prefer E-112 wherever the point is refusal. E-010 and E-011 are Untested — citations unresolved since the C-017 triage — and the beat rests on the HMDA rows, not on them. E-014 usefully complicates the clean legal story: 60% of these borrowers own their land and 17% still take chattel, so channel matters, not just statute.
The lender refuses on the classification, not the borrower. Watch what that does to the market.
373,143 shipped in 1998. 130,748 by 2004 — while site-built housing was booming.
The peak was 1998: 373,143 homes. By 2004 it was 130,748 — a 65% fall, reached before the financial crisis, in the years site-built housing was having its biggest boom in decades. The floor came later, 49,717 in 2009, 87% below peak. Read as a share it is the same story: manufactured homes were about a fifth of new single-family houses before 2000 and roughly nine percent today. Shipments have since recovered above 100,000 — 112,882 in 2022, 102,738 in 2025 — and are still under a third of where they were. The competing explanation. Anyone who knows this sector says the collapse was the subprime bust — Green Tree, bought by Conseco in 1998, and the securitized lending that failed with it. Conceded: the credit event is real and lands exactly on these years. But it was a chattel credit event, and chattel lending is what the vehicle title produces. A house financed as a mortgage was not in that market to collapse out of. A second federal series is quoted as saying the same thing from a different angle — 332,000 new manufactured homes sold in 1996 against 118,000 in 2005, a 64.5% fall on a sold series rather than a shipped one, which is closer to demand. This project has not opened the GAO report those figures come from, so the row carries them Untested with the citation unresolved (E-130). The collapse does not rest on them; E-012 is read from the Census table directly. The stronger form of the objection is that repossessed homes, not withdrawn credit, did the damage — and it is larger than we first said. Two lenders' own filings: Conseco Finance held 24,131 unsold repossessed or foreclosed properties at the end of 2001 and liquidated 25,750 units that year at 57% loss severity; Oakwood repossessed 13,423 in fiscal 2002. Oakwood's filing puts the industry at 90,000 repossessions in 2002, more than half of new home shipments. And that is the same argument one layer down. A home titled as personal property is repossessed under UCC §9-609 — "without judicial process, if it proceeds without breach of the peace". A mortgaged house requires foreclosure. What the title changes is the cost and time to clear each unit, not the size a wave can reach: foreclosure moved millions of site-built homes after 2008, so capacity was never the point. §9-609 takes the court out of every individual repossession, and foreclosure does not. Competing supply is not an alternative to the classification argument. It is a second consequence of it. The chain, stated so it can be attacked — and it forks. Two links in series, then two branches: One — the title. The 1974 Act preempts construction standards and is silent on property classification, so states classify through motor vehicle agencies, and three-quarters have written statutes to convert out of it (E-091, E-092, E-120). Two — the financing. Personal property means chattel paper rather than a mortgage (E-013, E-011). Then it splits. Branch A — the money withdrew. Chattel paper carries higher rates and shorter terms, and when the securitisation market for it failed, originations went with it: securitisation halved in 2000 while shipments fell 28.1%, financing contracting 1.8 times as fast as the volume it financed (E-122). Branch B — the inventory arrived. Chattel collateral is recovered under §9-609 without a court, per unit, cheaply (E-124), and the recovered units came back onto the market at a scale that competed with new production. Conseco liquidated 25,750 in 2001 at 57% average loss severity — 43 cents on the dollar, a fire sale in the company's own income statement — while holding 24,131 more unsold. The industry estimate is 90,000 in 2002, over half of new shipments (E-129, E-123). Both branches end at the collapse: 373,143 shipments in 1998, 130,748 by 2004, 49,717 in 2009 (E-012). Why it is drawn as branches and not a line. In series, the weakest link governs the whole argument, and the weakest link here is the per-unit repossession claim — the mechanism is in the statute and the magnitude is measured nowhere. Drawn honestly, that weakness costs Branch B and leaves Branch A standing on the securitisation-versus-shipments evidence, which does not depend on it at all. A chain drawn in series would have overstated what a single doubt destroys.
InvestorsThe obvious objection to a 1998-to-2009 chart is that 2009 was the financial crisis. It does not survive the intervening years: two thirds of the decline had already happened by 2004, while every other kind of housing was booming. Whatever did this was specific to the product, and it was not construction cost.
PolicymakersThe number for anyone who believes the supply problem is fundamentally land or labour: a product with a federal standard, a large price advantage and no zoning fight at the factory door lost two thirds of its volume during a housing boom. And the subprime bust does not explain it away — that was a chattel credit event, and chattel lending is what the vehicle title produces (E-122).
ResearchTwo channels, separated only partly. One discriminating fact: securitization of chattel loans halved in 2000 while Census shipments fell 28.1% — 1.8 times as fast, an origination-led collapse (E-122). That ratio uses our own denominator. The report's looser 20 percent would make it 2.5, and taking the flattering version of your own argument is how a case gets overstated a fraction at a time (C-025). E-124 establishes the legal mechanism, not the realised speed. No measured comparison of repossession-to-resale against foreclosure-to-resale exists. The claim is what the statute permits, not a measurement of what happened. E-012 is the US total row of the Census annual shipments-to-states table. It replaced a figure sourced to "Urban Institute", then was replaced again: the first reading scraped the monthly workbooks, mis-assigned years, and produced the claim that shipments had not been above 100,000 since 1998. They were, four times in the last five years — C-024. A derived series is a claim. Census warns pre-2014 data is not comparable; 1998, 2004 and 2009 all sit before that break, and 2026 is excluded as partial.
A collapse that runs opposite to the cycle has a mechanism, and the mechanism outlived the thing that caused it.
Congress struck the permanent chassis from the federal definition on 11 July 2026. The classification it was used to justify lives in state law, and the Act reached that too.
The 21st Century ROAD to Housing Act took effect on 11 July 2026. Section 301 amends the statutory definition of a manufactured home by striking "on a permanent chassis" and inserting "with or without a permanent chassis". The physical premise of the vehicle classification — that this is a thing built to be towed — is repealed. What that section does not do is reclassify anything. The classification itself lives in state law, in motor vehicle agencies, and in conversion statutes that three-quarters of states have written some version of (E-120). A definition in federal law changed; fifty titling regimes did not.
InvestorsThe famous half of this Act is the smaller half. Removing five words from a definition does not make a home financeable, and nobody has published a saving for an ordinary single-storey home. What matters is that the same section did not stop there.
PolicymakersThis is the actionable finding of the entire Act, and it belongs to legislators rather than to anyone building software. The federal justification for titling a house as a vehicle is gone as of July 2026. The state statutes that do the titling are still there, and the instrument for changing them already exists — a conversion statute, which most states have.
ResearchThis site said for five weeks that the consequences of the repeal were "not scheduled". That was wrong, and it was wrong because E-090 declined to claim the rest of the Act as unverified and a beat then wrote a sentence that was only true if the unverified part was absent. An unopened source is not an empty one. Correction C-046; the record of what was read instead is E-197 through E-202.
It reached it with a date attached. That date is the only thing in this argument that expires.
The same section obliges every state to certify by 11 July 2027 that its own laws treat a chassis-free home identically — financing, title, insurance and taxes named expressly.
Section 301(c) added a new subsection to the same statute the definition sits in. Under 42 U.S.C. 5403(i), not later than 11 July 2027, every state must certify to the Secretary that its laws "subject a manufactured home without a permanent chassis to the same laws and regulations of the State as a manufactured home built on a permanent chassis, including with respect to financing, title, insurance, manufacture, sale, taxes, transportation, installation". Two years is allowed where a legislature meets biennially. Recertification is annual (E-197). Read that list against Act III. Financing, title, insurance, taxes are the questions the 1974 Act left to fifty states, and they are the questions this Act now puts on a dated federal filing. One more provision, in the same title and unremarked: federal home-improvement lending now names the accessory dwelling unit as an eligible use (E-212). The instrument that was missing for a casita on a lot someone already owns is the one that moved.
InvestorsThis is the only thing in this argument with an expiry date. Eleven months in which fifty legislatures must pass a bill about title, financing, insurance and taxes for a product class that does not physically exist yet — and no model certification language has been published by anyone. Everything else on this site can be measured next year at the same price.
PolicymakersThe clock is real and the pen is unheld. A state's choices here are two lines of drafting apart and they are not equivalent, which is the next beat but one. What a legislature will reach for by default is whatever its own counsel drafts fastest in a session with fifty other bills on it.
ResearchRead from the codified text, subsection by subsection, because the secondary briefs describe this provision and do not quote it. Subsection (i)(4) is the part with research value: the Secretary "shall publish and maintain in the Federal Register and on the website of the Department of Housing and Urban Development a list of States that are up to date with the submission of initial and subsequent certifications". That makes compliance a published series rather than a records request — the first instrumented before-and-after this project has been able to name, on a schedule set by statute.
A deadline is only a deadline if something happens when it passes. Something does.
Non-certification does not lapse quietly. It obliges the state to prohibit the manufacture, installation or sale of the chassis-free home.
If a state does not certify by its deadline, 42 U.S.C. 5403(i)(5)(B) requires that "the State shall prohibit the manufacture, installation, or sale of a covered manufactured home within the State" — a covered home being one built without a permanent chassis after 11 July 2026. Where the Secretary administers installation, state and Secretary both prohibit it. The Secretary may not waive the prohibition for a late certification unless the Secretary approves that certification (E-198). The enforcement runs against the house, not against the state.
InvestorsRead plainly, this is the one provision on this site that could make the addressable market smaller than it was before the Act. A statute passed to widen supply forecloses the new product in every state that misses its own deadline, and the relief is discretionary.
PolicymakersWhatever the drafting intent, the operative effect of a missed deadline is a prohibition on a class of affordable housing inside the state that missed it. A legislature that does nothing is not holding position; it is voting for the ban by default, and it will not read that way in the session.
ResearchHow many states certify is E-200, and it is empty — but it is the cheapest open row on this site. The denominator is fifty, the deadline is statutory, and the Secretary is required to publish the list. This is the rare gap that closes by reading rather than by fieldwork, which is exactly the criticism C-048 earned on E-037.
So the deadline binds. The question is what a state is actually required to write by it.
The certification demands parity, not reclassification — and the parity runs to whatever the state already calls a manufactured home.
The obligation is parity, and the parity has a referent: a state certifies that its laws "treat any manufactured home in parity with a manufactured home (as defined and regulated by the State)". Where the certification is not filed with a state plan, the statute names a compliance route in its own text — an attestation that the state has taken the necessary steps, "including, as necessary, by amending the definition of 'manufactured home' in the laws and regulations of the State" (E-199). For most states that is six words added to a motor-vehicle titling statute. The certification is satisfied. And the classification the beats above describe is not merely preserved — it is re-enacted, in a fresh statute, with a federal filing behind it. The inference, stated so it can be attacked. A classification held in place by 1974 silence is easier to unwind than one a legislature affirmed in 2027. Nothing in the statute says states will take the cheap route; the statute permits both and names the cheap one first. Which route fifty states take has not happened yet, and it is the whole question.
InvestorsThe same provision is the best thing that has happened to this argument and the most likely way it dies. Eleven months of drafting decide which, and the drafting is unfunded, unassigned and mostly unnoticed.
PolicymakersA state has two routes to the same certification. It can extend its existing definition, which is fast, costs nothing and locks the chattel treatment in by statute. Or it can default a chassis-free home to real property, which most states have the machinery for already — three-quarters have written a conversion statute (E-120), and a conversion statute is a procedure for doing once what a definition could do by default. Both routes certify. Only one of them changes how the house is financed.
ResearchBoth quotations are the codified text. The consequence is not: it is an inference from what legislatures do under deadline, and it carries no evidence. E-200 is opened to hold the measurement that settles it — not how many states certify, which will be most of them, but how each one does, which nobody is recording. If certifications reclassify without anyone drafting anything, this beat is wrong in the direction this project would most like to be wrong in.
That is the argument's best chance and its likeliest death, in one provision. What decides which is drafting nobody is doing.
The paragraph directing HUD to write the standard a chassis-free home is built to carries no deadline at all — so fifty legislatures will legislate about a house nobody can yet build.
Section 301(b) directs the Secretary to issue revised standards for homes built without a permanent chassis, through the consensus-committee process. No date appears in the paragraph. That committee met twelve days after enactment and carried one substantive item: energy standards (E-201). The only chassis rule HUD has proposed predates the Act, does not mention it, and is narrower than it — upper floors only, priced on an assumed twenty to twenty-five multistory homes a year. Comments closed on 11 August 2026; no final action is projected (E-203). Meanwhile the statute requires the chassis-free home to carry a distinct label and data plate (E-202), and both Enterprise guides still define an eligible manufactured home as one "built on a permanent chassis" (E-204). A home the Act now permits would arrive carrying a label the rulebooks do not recognise (E-205).
InvestorsTwelve months for the states, no months for the agency, and a rulebook amendment nobody has scheduled sitting between the product and a mortgage. None of the three is a research problem. Four named institutions could each resolve their part in a paragraph, and the question of whether they will is E-206.
PolicymakersThe sequencing is backwards and it is fixable. States will draft parity language in 2027 for a home whose federal standard, label and secondary-market treatment do not yet exist — so the drafting has to anticipate all three or it will certify the wrong thing. A model bill is cheaper than fifty repairs.
ResearchOne row here cuts against this project and is kept for that reason. E-207: for a home certified to MH Advantage or CHOICEHome standards, site-built comparable sales are already required rather than merely permitted, in three rulebooks, one of them since 2023. This site has argued the comparables lock is a lock nothing has opened. For that class it is open, and correction C-050 narrows the claim to what survives — valuation by type, a certified design carrying its own value independent of location, which none of those provisions does.
A statute that lands in a system with no clock on the other side is not a new problem. It is the oldest one here, and four attempts have already run into it.
Movement 5 of 5
Four attempts, each failing bigger than the last
The technology worked. The demand floor was a government appropriation — and appropriations get cancelled.
FIXEDFactory production, at federal scale
KILLED BYThe demand floor was an appropriation — withdrawn in 1973
Twenty-two housing system producers selected from over two hundred competitors. The manufacturing half largely worked. The market-aggregation half never did, and the 1973 termination of federal production subsidies removed the demand floor participants had underwritten against.
InvestorsBuilt the factory and bought the order book with a subsidy. The order book was an appropriation, and it was withdrawn.
PolicymakersThe program is remembered as a technology failure. It was a demand-aggregation failure, caused by a budget decision three years in.
ResearchDiagnosed as an engineering post-mortem: built the production layer, attempted demand aggregation via subsidy, left the approval, information and valuation layers untouched.
So maybe the missing input was money. The next attempt had more of it than anyone, ever.
Billions of dollars. No capital constraint at any point. Bankrupt anyway.
FIXEDVertical integration of design and manufacture
KILLED BYA manual review queue upstream of everything it built
Vertical integration of design and manufacture, funded at a scale no prior attempt enjoyed. It optimized the build step while the pipeline around it stayed manual. A faster compiler behind a manual review queue produces no throughput.
InvestorsThis is the slide that should make an investor uncomfortable, and it should. The most-funded attempt failed without ever facing a funding problem. Money was not the missing input.
PolicymakersNo amount of private capital routed around the approval process. It waited in the same queue as everyone else.
ResearchRead as a controlled experiment: capital and manufacturing technology were held abundant, and the outcome did not change. That isolates the binding constraint to somewhere else in the stack.
So not capital either. Which leaves the rules themselves, and what nobody does to them.
A rule is compiled when a computer can check it.
Prose
"Compatible with neighborhood character."
Read by a person. Interpreted. Arguable. One project at a time, in each of ~20,000 jurisdictions.
Compiled
setback.rear >= 15ft
Checked by a machine. Deterministic. Reproducible. Free at the millionth query.
The way a tax form computes a number, not the way an essay is graded. America's building and zoning rules are written as prose, so nothing checks them automatically. Every project is read, interpreted and approved by hand, one at a time.
InvestorsProse is why nothing carries forward between projects, and why housing has no products — only projects.
PolicymakersThis is why a reform can pass and change nothing on the ground. The statute is prose too, and it lands in a system that can only process prose by hand.
Research"Compilable" here means expressible as a deterministic rule over measurable parcel and design attributes. What share of American provisions meet that test has never been measured — see the objections beat.
That is the definition. The record of applying it at scale — thirteen jurisdictions, one code — is the strongest evidence in this act.
The cost of encoding fell. What decides whether anyone adopts this was never mostly technical.
Turning prose rules into executable logic used to mean a person writing out each provision by hand, and language models have moved that cost for the first time. This site used to claim the cost was what closed the market. The best available study of why automated compliance checking gets adopted, across eight countries, finds twelve variables governing adoption and eight of them non-technical — government support, industry readiness, human resistance, education and training among them. That is the finding, and it is the part that carries. The transplants illustrate it. Singapore's engine was taken up in Norway and in a completed New York City pilot on ICC codes, with testing on Japanese and Australian models. All four inherited the encoding, so for those four the encoding cost was zero, and none reached production. That does not prove encoding cost was never binding anywhere. It removes it as the explanation for those four, and it leaves this: no case has been identified in which cheap encoding produced adoption.
InvestorsThe honest position is smaller than the one this site used to state, and more useful. A cost fell; no market opened because of it. What the evidence points at is a set of non-technical conditions — who assents, who is accountable, what the industry is ready to do — things a business arranges rather than computes.
PolicymakersThe American thread has a named cause, and it complicates us rather than helping. The ICC built SMARTCodes in 2006 to make its model codes machine-readable; development ended in 2010 for lack of funding. Fiatech's AutoCodes followed and stalled after its 2012 proof of concept. A project that dies in a budget is contingent, not structural — it says nothing about encoding cost or about assent, and we do not get to count it as evidence for our diagnosis. And the finding that cuts the other way. The objection to compiled review is that it sacrifices judgement — yet when thirteen jurisdictions reviewed the same plan set against the code's most objective chapters, the judgement being protected produced a reported spread of one flagged issue to 43. The status quo is not a reliable process automation might degrade; it is an unreliable one automation would stabilise.
ResearchSort the failures by what is known; they are not the same kind of fact. Named cause: SMARTCodes, funding, 2010 — peer-reviewed and independently corroborated. Assessed as unexplained: CORENET, which Potter — conceded to on prefabrication cost — says "died silently, so the reasons are unclear". Absence only: Norway, New York City, Japan, Australia. The New York City claim was challenged in review as a suspected conflation with SMARTCodes. It held — Khemlani (2005) reports a completed NYC pilot on ICC codes, and the chronology rules the conflation out — and its corroboration gap is recorded rather than glossed: one trade article, and three later surveys of this field that do not mention it. Two cautions on the spread. It is reported, not retrieved — ENR 403s and the Fiatech report is gone from the web and the archive — so E-126 stays Supported. And consistency is not correctness: an encoding picks one reading and applies it everywhere, which beats thirteen readings only if the reading is right, and somebody must be authorised to pick it. The assent question, with a number attached.
So the question is not whether rules can be compiled. It is what the attempts that tried actually died of.
Four attempts. Each fixed one layer, left the rest manual, and was killed by a layer it did not touch.
Four attempts. Each turned one lock. Each was killed by a lock it never touched.
Breakthrough built production and bought demand with a subsidy. CORENET compiled the rules for one jurisdiction by hand. Katerra industrialized the build behind a manual queue. SMARTcodes and AutoCodes compiled the model code itself; adoption stalled anyway, on grounds never mostly technical. The pattern is not bad luck — it is a stack treated as a single problem. The largest case you have already read: a federal standard that turned one lock at national scale, strangled by the one it left.
InvestorsFour different failure modes, one shared cause: partial coverage of a system whose locks are complements, not substitutes.
PolicymakersEvery one of these was a serious, well-resourced, competent effort. Treating them as blunders misses the finding entirely.
ResearchThe pattern is derived from the American record. Whether it reproduces internationally is the next beat — it does, and that is stronger evidence than one national record.
That is the American record. Does it reproduce elsewhere?
Tokyo, Auckland, Vienna, Singapore. Each solved a genuinely different constraint and left the others in place.
| City | Approval | Land | Finance | Demand | Cost |
|---|---|---|---|---|---|
| Tokyonational zoning | ● | ○ | ○ | ○ | ◐ |
| Aucklandunitary plan | ● | ◐ | ○ | ○ | ○ |
| Viennasocial housing | ○ | ● | ● | ◐ | ○ |
| SingaporeHDB + CORENET | ● | ● | ● | ◐ | ○ |
| Arizonacasita preemption | ◐ | ○ | ○ | ○ | ○ |
| United Statescurrent state | ○ | ○ | ○ | ○ | ○ |
● solved · ◐ partially addressed · ○ not addressed
Tokyo: twelve national zones, residential permitted in all but one, as-of-right approval. Auckland: three-quarters of residential land upzoned, rents 23% below the synthetic-control counterfactual at eight years. Vienna: about 60% of residents in public or limited-profit housing, land taken permanently out of the market. Singapore: land, finance and process together — via a state with compulsory acquisition powers no American jurisdiction possesses. Arizona: the right granted in 2026, and nothing built to exercise it.
InvestorsNote what nobody has: a compiled process layer in a jurisdiction that also granted the right. That is the whitespace.
PolicymakersAuckland is the important one for a legislator. It is the cleanest large-scale natural experiment available anywhere, and it settles the question of whether the saving reaches residents: consents up 21,808 at five years, permits per capita doubled against counterfactual, rents 23% below at eight years.
ResearchTwo caveats travel with this slide. Tokyo's permissive zoning is downstream of an asset structure America does not have — Japanese homes depreciate to near zero on a thirty-year rebuild cycle, so homeowners hold no supply-restriction interest. That claim is graded Supported, not Confirmed, because the causal direction is argued rather than established. And Vienna has a serious published critique: engage it rather than dismissing it.
It does — and the same gap is exposed every time. The locks are not independent either.
Politics chooses the ground. Information is the wedge. Capital is the thesis.
Three things block American housing and they bind in a strict order. Entry: can this be built here at all — politics decides. Throughput: given permission, how fast and how certainly does a unit move — information decides. Survival: does the product last at scale — capital decides. The binding constraint is not a fact about housing. It is a function of which stage you are standing in.
InvestorsSequence by politics, wedge with information, win on capital. Type-based valuation is the endgame because it is the lock that kills at scale, and the position with no identified competitors.
PolicymakersPolitics first does not mean lobbying first. It means letting the map choose the segment — build where the incumbent homeowner is also the builder, because that is the only ground where entry has already been granted.
ResearchNote the asymmetry in evidence strength. The valuation lock has the best evidence because it is the only lock America has ever gotten far enough to fail at. Evidence strength tracks how far each attempt got, not how much each lock matters.
Every other sequence has already been tried.
Four sequences, four failures, and each one is already in the record.
| Sequence | Who ran it | Why it failed |
|---|---|---|
| Capital first | Katerra | Money waited in a manual queue. No certified type for finance to attach to. |
| Information first | The screening market | Compiled rules without political standing are advisory. The rules got compiled and the approval did not move; how much time it took out has never been measured (E-060). |
| Politics only | Arizona | The right granted, the plans mandated, and the volume flat. No shared record, no type-based finance. |
| Politics + capital | The HUD Code | Closest anyone came. No information layer, so finance defaulted to chattel. |
Capital first: Katerra. Money waited in a manual queue. Information first: the screening market — Symbium, Buildability, Canibuild — compiled rules across thousands of counties and did not displace a single approval, because compiled information without political standing is advisory. How much time it took out has never been measured by anyone, including us (E-060). Politics only: Arizona, a granted right with nothing to exercise it. Politics plus capital, skipping information: the HUD Code, the closest anyone came, with no type-based valuation record to lend against.
InvestorsThe screening market matters most here. It is the information-first attempt at commercial scale, and the observable part is qualitative: the rules got compiled and the approval did not move. Compiling the rules is necessary and insufficient — the wedge has to carry legal weight, not just information. No number belongs in that sentence. This site used to put one there, sourced to an estimate of its own that it described as the market's published answer; the estimate is withdrawn and nothing replaces it (C-019).
PolicymakersThe failures are not symmetric. Skipping politics wastes a company. Skipping information wastes a statute. Skipping capital wastes an industry.
ResearchThis corrects an error in the thesis as published: Red Team 04 claims existing vendors leave the ruleset in prose. Register row E-040 refutes that. The sharper form is the one on this slide.
Every order that failed lacked the same instrument: a check certified once, whose result travels. One city turned a lock without it.
The most useful thing about Auckland is not its size. It is that somebody measured it.
The only place one lock has been turned and measured
That is one lock, turned once, in one city of 1.7 million. Three locks turned across twenty thousand jurisdictions has never been tried, cannot be extrapolated from a single case, and is not being claimed here. What Auckland establishes is narrower and more useful: that when supply actually responds, the saving reaches the people who live there rather than capitalizing entirely into land. That was the objection everyone raised, and it is the one that has been answered by measurement.
Auckland upzoned three-quarters of its residential land in 2016. Five years on, an additional 21,808 homes had been consented — about four percent of the housing stock — and permits per capita had doubled against the counterfactual. Eight years on, rents were twenty-three percent below where the synthetic control says they would have been.
InvestorsOne reform, one city, and the effect is large enough to see through the noise eight years later. That is the closest thing to a demonstration this field has.
PolicymakersThis is the answer to the objection that supply reform is swallowed by land values. It is not an argument; it is a measurement, from a jurisdiction that did the thing and then let economists check.
ResearchAnd it is one city of 1.7 million, turning one of three locks. It does not license extrapolation to twenty thousand American jurisdictions and none is offered here. What it establishes is narrower: that when supply genuinely responds, the saving reaches residents rather than capitalizing entirely into land.
That is what one lock, turned once, produces. And none of what would turn the rest exists today.
Movement 6 of 5
What was designed, what it is worth, and what would kill it
Not "faster." Different in kind — six properties. One of them this country already built, for one kind of house. The other five do not exist.
| Property | Today | Built |
|---|---|---|
| Compliance | Discovered at review, after land is committed | Computed before commitment |
| Parcel facts | Rebuilt independently by six parties | One record, many readers |
| Gates | Serial, no deadline, nobody owns the total | Concurrent, on a clock, defined default |
| Reuse | Nothing carries forward between projects | Certified types carry forward by reference |
| Valuation | Comparable sales — a new type cannot be valued | Attaches to the type; location adjusts |
| Production | Speculative; utilization cannot be planned | Against a committed forward book |
Nothing here changes what a building must achieve. The standards are unchanged — they become checkable.
Compliance moves from something discovered at review to something computed before commitment. Parcel facts stop being rebuilt by six parties and become one record with many readers. Gates stop running in series without a deadline. Designs stop being redrawn and start being referenced by version. Valuation stops depending on the accident of having neighbors who already bought one. Production stops being speculative.
InvestorsRow four is the one that compounds. "Nothing carries forward" is precisely what makes housing an industry of projects rather than products, and it is the row that turns a service business into a product business.
PolicymakersNote what is not on the list: nothing changes about what a building must achieve. The standards are untouched. What changes is that they become checkable before somebody has committed to land.
ResearchSix properties, six falsification tests. This is the argument's exposed surface — any row that cannot be delivered kills the claim for its layer, and the layer's substitutes are specified rather than assumed.
Five of the six are absent. The sixth is the precedent this country already runs, and it stopped where it started.
Housing already certifies a design once, delegates the check, and lets the result travel — as long as one standard governs. Take away the single standard and it stops at the state line.
| Aviation | Pharmaceuticals | Automotive | Housing, already | Housing, otherwise | |
|---|---|---|---|---|---|
| What gets certified | The design, once | The reference product | Nothing, before sale | The design, once — 1976 | Every project, anew |
| How each unit clears | Conforms to the type | Equivalent to the reference | The maker’s certificate | Conforms to the type | Re-reviewed by hand |
| Who verifies | Accredited organizations | The manufacturer | The maker, alone | A delegated or private agency | Every jurisdiction, alone |
| Cost of the tenth unit | A fraction of the first | A fraction of the first | A fraction of the first | A fraction of the first | The same as the first |
| Does it travel | Yes, nationally | Yes, nationally | Yes, all fifty states | No — one class, or one state | No |
Failure kills in all five columns. Housing already certifies a design once and delegates the check — federally since 1976, in Florida since 2006. Cars clear no agency at all, and the maker’s certificate is good in every state, backed by a recall housing has no version of. The row housing fails is not who verifies. It is whether the result travels.
Housing cannot ship and iterate. There is no rollback, no staging environment, no canary deploy, and a defect is not downtime but something structural, permanent and occasionally fatal — which is why "move fast and break things" is disqualifying here, and why it describes what went wrong at Katerra. Safety-critical engineering met that constraint by qualifying the toolchain rather than inspecting every output: certify the checker, and treat conforming output as verified by construction. Cars are the same move at a harder setting: no agency approves a vehicle design before sale, the maker certifies conformance itself, and because the standard is federal a single determination is good in all fifty states. What backs that is the recall, which housing has no version of — so the half worth taking is the portability, not the self-certification. It also changes what the designer can do — a configurator built against a certified envelope cannot express a design outside it, so non-compliance becomes something you cannot author rather than something you detect.
InvestorsThe same move exists in pharmaceuticals: abbreviated approval lets a product reach market on demonstrated equivalence to a reference rather than by repeating the trials. That is the template for the valuation lock — a housing type valued on equivalence rather than on the accident of having neighbors who already bought one.
PolicymakersThis answers the objection that skipping first-pass review is unsafe, and it is not an analogy — it is American practice. Delegated agencies have approved HUD-code designs since 1976; Florida has allowed private plan review in lieu of the building department since 2006. Nothing here proposes buildings go unchecked, only that the checking be done once, to the design, by a qualified checker. And it argues for something other than what this project proposes. A HUD-accepted agency's approval is good in any state (§3282.351(d)) because one federal standard preempts local rules. Portability belongs to the standard, not the determination — so the precedent says: preempt first.
ResearchThe tension cuts at the thesis, not at the objection. Every portable determination here rides on a preempting standard — HUD-code approval on 42 U.S.C. §5403(d) (E-091), a vehicle certificate on 49 U.S.C. §30103(b)(1) (E-111). Florida's private plan review has none behind it and does not travel (E-089). The record holds no case of a determination crossing a boundary where the receiving jurisdiction kept its own rules. The inference that invites — the missing instrument is a preempting siting standard — is graded Untested, what would settle it named (E-229). These are cited as regulatory structures, never as evidence that delegation improves outcomes — automotive for portability alone, its enforcement being post-sale, on a remedy housing lacks. Whether type certification extends to site-built housing is untested.
Already ours — and every case that travels rides on preemption; the close now states that implication, graded. A standard stated plainly enough to be met has an obvious borrowed form.
Commercial real estate says Class A, B or C. Residential says three bedrooms and a good neighbourhood — and leaves the standard unstated.
Commercial property carries a letter. BOMA calls the A/B/C system "a subjective quality rating of buildings which indicates the competitive ability of each building to attract similar types of tenants" — explicitly a guideline rather than a certification, and relative to its market: Class A in Manhattan is not Class A in Winnipeg. It works anyway, because it is anchored to something measurable. A class predicts rent. Residential has no such vocabulary. It describes — square footage, bedrooms, the neighbourhood — and leaves the judgement to whoever is reading. That is also, roughly, what discretionary review does: a standard gets applied without being stated, and a project is refused for not fitting a community whose criteria were never written down. A published grade with explicit features would put those criteria where they can be met, argued with, or compiled.
InvestorsThe commercial letter exists because income made it worth writing. Residential has no equivalent anchor, which is the honest objection to this and the thing that would have to be built first: the letters are downstream of the thing they predict, not a substitute for it.
PolicymakersThis is the part of the argument a jurisdiction could act on without waiting for anything else on this site. Publishing what a community actually requires, as explicit features rather than as prose applied case by case, is a smaller act than adopting an instrument — and it converts discretion into a standard that can be met rather than guessed at.
ResearchUntested, and the gap is specific. E-039 — the objective-versus-discretionary split of American zoning — has never been measured anywhere, so the claim that discretion is used to gate projects on implicit standards is an argument here, not a finding. E-216 is Supported only for what BOMA says about its own system, quoted through industry guidance rather than from BOMA directly. Nobody has tested whether a residential grade would carry information that square footage, bedroom count and location do not already carry.
A grade only means something if somebody stands behind it. That is the missing office.
A conformance authority — designed in full, deferred in full, and revivable by two measurements.
What was designed is a conformance authority called Lintel: five stages, only the first three new — compiled rules, a determination against one parcel, a professional signature with insurance behind it — and the check itself certified rather than each building (E-088). It would answer two questions. One is advisory and already sold by others: send an address, get an opinion (E-040–E-042). The other carries legal weight and nobody sells it: send a certified type and a parcel, get a determination a jurisdiction accepts in place of first-pass review. Seven layers, each with its own kill test; three ordinary building types that are legal somewhere and impossible here (E-036). All of it is deferred, because the two joints that carry the load are unmeasured: whether binding siting provisions are mostly computable (E-039), and whether the product is cheaper like-for-like (E-098). Those two measurements would revive it. Until then this section states the design and does not argue it.
InvestorsThe distinction that matters: screening tools produce opinions, the jurisdiction still performs its own review, and what that market did to the clock has never been measured (E-060). A determination would carry a professional obligation and an insurer behind it. That trade — advisory to load-bearing — is the product, and it is exactly the part no jurisdiction has accepted.
PolicymakersNothing here asks a jurisdiction to surrender judgement: discretionary provisions stay with the jurisdiction, referred by construction. What the design removes is re-verification of the already-certified — the trade Arizona already makes for the factory-built shell itself (A.R.S. §41-4006, E-132).
ResearchThe design documents are HAP-05 and HAP-06, and the sixteen-model autopsy is on /models. The rows this section rests on keep their grades: E-039 and E-098 empty, E-060 unmeasured, E-047–E-050 as graded. Nothing in this section upgrades them.
Complete on paper, and never run. Whether the rules beneath it can be computed at all is the prior question.
The chart is schematic. The crossing point has never been measured — and that is the claim, not a weakness in it.
Stated plainly, at the moment it would pay not to
The shape of the curve is the claim.
The axis values are the research question.
Quantifying the crossing point — cost per jurisdiction to encode, at a stated accuracy level — is the single most valuable number this project could produce, and it does not currently exist.
The shape of that curve is the claim; the axis values are the research question. Cost per jurisdiction to encode, at a stated accuracy, is the most valuable number this project could produce, and it does not exist. The method has changed: hand-encoding was the only way to turn prose into executable rules until about 2023 (E-051), and large language models make that step something a machine can attempt. That moves the cost, not the question — four jurisdictions inherited Singapore’s encoding free and none reached production (E-080, E-081).
InvestorsAn opportunity that is real and unmeasured is the only kind still available to a new entrant. If the number had been published, the position would already be taken.
PolicymakersNo state has instrumented any of this. Arizona passed a significant housing reform and built nothing to determine whether it worked. Confirming that absence is itself a finding.
ResearchThirty-two register rows are empty and they are the ones the argument leans on hardest: E-039, the objective-versus-discretionary split, never measured. E-051, the cost to compile one jurisdiction. E-037, Arizona permit volume since preemption. E-038, the idea-to-sale cycle. E-060, what compiled screening does to cycle time. Plus E-070 and E-071, both assumptions.
Unmeasured is the biggest hole in this argument. What a compiled path would do to the clock is next — modelled, and only modelled.
The same house, on a different clock.
Today · 32 months
Compiled · 7 months
Query and certify in under a month. Two months of site preparation. Four months of factory build and set. That is twenty-five months off the clock, and the split matters more than the total: about twenty-one of them are approval and information processing, and about four come from building in a factory rather than on site. Only the twenty-one belong to compiling the rules. The rest is a change of construction method, and crediting it to approval is the error that took this beat off the site once already. Every stage that compresses to near zero is an information-processing stage. Every stage that resists compression is physical.
InvestorsFour prior attempts to industrialize housing concentrated on the one row of that table that is hardest to compress, and left the six easiest untouched.
PolicymakersThe argument is not that houses should be built faster than is physically sensible. It is that the twenty-five months surrounding the build are overhead, and they are administrative overhead.
ResearchThe compiled column is E-070 — modelled, never run, graded Untested — not a measurement, and this follows if — and only if — the pathologies at beat 07 are actually fixed. Some stages overlap in practice; this is critical path, not the sum of all activity.
Twenty-five months gone. Why does that matter so much?
The same margin, on a different clock, unlocking new capital.
Below the hurdle rate of almost every risk-capital pool.
Identical margin, identical product. Only the clock changed.
The consequential number. Double today's return at 40% of the margin.
On the modelled clock — E-070, never run — a 25% margin over 32 months annualizes to about 9% — below the hurdle rate of almost every risk-capital pool. The identical margin over 7 months is about 47%. And a 10% margin over 7 months is about 18%: double today's return at 40% of the margin.
InvestorsHousing is not underfunded because investors doubt the market. It is underfunded because the learning cycle is longer than a fund's life. Software invalidates a hypothesis in twelve to eighteen months for a couple of million dollars; housing takes three to five years and tens of millions. No amount of conviction fixes arithmetic about how risk capital forms.
PolicymakersRead the third card. Speed converts directly into a lower price without anyone accepting a worse return. That is the mechanism by which a process reform reaches a household — and it is why the argument is about time rather than subsidy.
ResearchIllustrative calculations on stated assumptions, not measured outcomes: E-070 and E-071 are both graded Untested and the seven-month clock has never been run. One outside model points the same way with published inputs: NAHB's schedule marks a dollar spent applying for development approval up 30.17% by closing, against nothing at closing (E-220) — arithmetic on stated assumptions, not an observation. Of the twenty-five months, only about twenty-one are approval; four are the factory. Gross margin on total project cost; ignores leverage, which amplifies the effect in both directions. And the argument is specifically about iteration rate, not capital volume — a version of this that argues for more money rather than faster learning is arguing for the wrong thing.
So what is all of that actually worth, and to whom?
If the modelled clock held — and it has never been run — here is the per-unit arithmetic, and the one measured precedent that a saving reaches residents.
Regulation is 26.4% of a new home's price — the pool the saving comes out of. Auckland rents came in 23% below counterfactual at eight years — measured evidence it reaches residents. No market size is claimed, because none has been estimated, and no builder or buyer figure is claimed either — both rest on rows E-070 and E-071, which are graded Untested and which compare site-built today with factory-built tomorrow — about four of the twenty-five months belong to the factory rather than to approval. Auckland is the one figure here with a measurement behind it.
None of this is an outcome figure, and the site claims none. The clock below is modelled and untested. Regulation is roughly a quarter of the price of a new American home, and that is the pool any saving comes out of. On a seven-month cycle a ten percent margin still returns about eighteen percent — double what a twenty-five percent margin returns over thirty-two months. Speed converts directly into a lower price without anyone accepting a worse return. And Auckland is the measured precedent that the saving reaches residents rather than capitalizing entirely into land.
InvestorsAn identical margin on a different clock is a different asset class — nine percent against forty-seven. That is the difference between an asset only patient real-estate capital will hold and one a portfolio strategy can underwrite.
PolicymakersThe reason to care is the third figure. Auckland is a supply reform that measurably reached renters at scale — rents twenty-three percent below the counterfactual at eight years. That is the outcome housing policy usually tries and fails to buy directly.
ResearchThe first figure comes from a trade association advocating deregulation and is graded accordingly. The bars are illustrative arithmetic on stated assumptions — E-070 and E-071, both Untested — and about four of the twenty-five months belong to factory build, not to approval. Auckland is the one row here with evidence behind it, and it covers upzoning rather than this. No market size is claimed.
All of which is worth nothing unless the people who must act actually want it.
The same house, in the system that works — and every piece of it already exists somewhere.
A walkthrough of a system that does not exist, from the developer’s seat. Every component named below exists somewhere; none of them exists together. You open a map and click a parcel. The buildable envelope comes back in seconds — deterministic, reproducible, free — because the rules governing it are executable rather than prose. You choose a certified type that fits inside it, with published cost and lead time. There is nothing to entitle, because conformance is verified rather than re-reviewed. The remaining checks run at the same time against one shared record, each on a statutory clock. The lender underwrites the type's record instead of the neighbours who happened to sell. The factory already had the order before it started.
InvestorsEvery arrow in that sequence is a place where money currently waits. Removing the waiting is the product — not building faster, which is the one thing here nobody is proposing to do.
PolicymakersRead what is absent from that sequence: no safety check was removed. Every review still happens. They happen concurrently, against shared facts, on a deadline.
ResearchEach step corresponds to a layer with its own falsification test. If the parcel record proves unobtainable, or the type cannot be certified, or the clock proves politically unenforceable, that step fails and its layer is replaced — the architecture is built so a failed layer is survivable.
That is how it would run from the developer’s seat. The artifact at the centre of it is worth seeing whole.
The product is a document. This is what it would have to contain.
Specimen — no determination like this has been issued. It is drawn to show what one would have to contain: what was checked, what was computed, what was referred to a human, who signed it, who insured it, and the statutory basis on which a jurisdiction would accept it. The counts are illustrative. The share of provisions that are objective rather than discretionary is register row E-039, and it is empty — which is exactly what Gate 0 measures.
A determination names the parcel, the certified type and the exact ruleset and edition it was evaluated against. It reports how many binding provisions were examined, how many were objective and computed, and how many required judgement and were referred to the jurisdiction. It carries a professional's signature, a named insurer, and the statutory basis on which it is accepted. It is auditable after the fact, because the ruleset version and the engine version are both on it.
InvestorsEverything that makes this defensible is on the face of the document: what was checked, what was not, who is liable, and who pays if it is wrong. That is what makes it insurable, and insurability is what makes it acceptable.
PolicymakersThe line that should reassure a building official is the third one. Sixteen provisions referred, reviewed as normal. The determination does not claim the discretionary questions — it clears the ones that are arithmetic so the reviewer's time goes where judgement is actually required.
ResearchSpecimen only. No determination like this has been issued, and the counts on it are illustrative rather than measured. The ratio it implies — how many binding provisions are objective rather than discretionary — is register row E-039, which is empty, and is precisely what Gate 0 exists to measure. If that ratio is bad, this document is thin and the argument is wrong.
That is what one page of it would say. The same walk, from the seat of somebody who just needs somewhere to live, reads differently.
Everything so far has been told from the seat of someone building. This is the seat of someone trying to get housed.
| If you need somewhere to live | Today | Built |
|---|---|---|
| Finding out what is possible | Months of professional interpretation, paid for, and the answer is an opinion | A query. Free, instant, the same answer for everyone |
| Knowing the price | Discovered by bidding against scarcity | Quoted before you commit, like any other made thing |
| Borrowing against it | A new kind of home: the loan is refused 42% of the time against 7% for site-built, at 5.61% spreads against 1.20% | A mortgage on a type with a published record — not a chattel loan on personal property |
| When it goes wrong | Litigation — slow, expensive, rarely proportionate | A warranty against the certificate holder |
| How long you wait | Thirty-two months, of which eight are anyone building anything | No figure is claimed. The comparison that stood here was withdrawn in August 2026 as not like for like — correction C-013 |
Every figure here is in the register, and where a figure was withdrawn the cell says so rather than carrying a replacement.
The same walkthrough, from the seat of someone trying to get housed rather than someone building. It describes a deferred design, not a service anyone can use. Today you cannot find out what is possible on a piece of land without paying for months of professional interpretation, and the answer comes back as an opinion. You discover the price by bidding against scarcity. If you want a new kind of home, your loan is approved under thirty percent of the time against over seventy for site-built, at more than four times the rate spread. If it is defective you sue. Built: the query is free, the price is quoted before you commit, the loan is a mortgage against a type with a published record, and the recourse is a warranty.
InvestorsThe third row is the business. Every other row is a better experience; that one is a different asset class, and it is the lock nobody is working on.
PolicymakersThis is the row of the argument that a constituent recognises. Not cycle time, not conformance determinations — whether they can find out what is possible, what it costs, and whether anyone will lend against it.
ResearchEvery figure here is already in the register and none of it is new. What is new is the seat: the argument has been made from the developer's chair throughout, which is a real weakness in how this case is usually put, and it is worth noticing that it took this long to correct.
That is the experience it would be. Behind it there would have to be a machine, and it is worth being exact about how little of it exists.
Three inputs, two programs, two signatures, three consumers — and a loop that makes the registry better with volume.
Nothing here is running. This is the design of record — HAP-05, HAP-06 and /models — for something the argument defers, and the crossing point it would produce has never been measured. Adopted codes, parcel records and a registry of certified types go in. A compiler turns prose provisions into executable rules pinned to a code edition; an engine evaluates a type against a parcel against that ruleset, deterministically. A professional of record signs the result and an insurer prices the risk of its being wrong. What comes out is a determination that a jurisdiction accepts in place of first-pass review, a lender values against, and a factory builds to. Outcomes flow back — cycle times, accuracy, resubmittals, performance in service.
InvestorsThe feedback arrow is where the durable advantage sits, not the compiler. Anyone can compile a ruleset. The registry that improves with volume, and the accuracy record that makes insurance cheaper each year, are what compound.
PolicymakersTwo boxes in this diagram are what make it acceptable to a jurisdiction rather than merely useful: a named professional who signs, and a named carrier who pays if it is wrong. Software alone would be advisory. Software plus liability is a determination.
ResearchThe compiler box is the one carrying the most weight and the least evidence. It presumes provisions can be pinned to an edition and evaluated deterministically at scale — register row E-051 for the cost of doing it, E-039 for the share that can be done at all. Both are empty.
That is the design of record, deferred in full. What the record around it implies — the sentence this argument has so far stopped short of — comes first.
Construction travels because one standard governs it. Siting travels nowhere, because none does. If the pattern in the record holds, siting needs what construction got in 1976.
Everything in this record that travels, travels the same way. A HUD-code design approval is good in any state because one federal standard preempts local construction rules (E-088, E-091). A vehicle maker’s certificate is good in fifty states because the standard behind it is federal (E-111). Florida’s private plan review — lawful since 2006, insured and clock-bound — has no preempting standard behind it and does not travel (E-089). Arizona draws the boundary inside a single section of law: construction preempted, siting expressly local (E-132). So the conditional, stated as an implication of those rows and nothing more: if a certified home’s right to land is ever to travel the way its right to exist already does, some part of siting has to come under a preempting standard. That sentence is graded Untested in this project’s own register (E-229) — the record holds no counterexample, and no case that settles it either. What has to give is not named because it is welcome; it is named so it can be tested.
InvestorsThe conditional is the investable form of the diagnosis: the silent months are priced where standards fragment, and the one place fragmentation was removed is the one place a determination has value across state lines. Nothing here says the fix will work — E-229 is Untested and the outcome rows are empty. What it says is where the experiment has to run.
PolicymakersThe clock for that experiment already exists and nobody had to propose it. 42 U.S.C. §5403(i) puts every state on a one-to-two-year deadline — 11 July 2027 for most legislatures — to certify that its own law treats a chassis-free home identically, naming title, financing, taxes and installation expressly (E-197). Fifty legislatures are about to open exactly the statutes this argument is about, and the compliance record will be a published series rather than a records request. A model act for that filing is drafted and public on this site.
ResearchThe inference is stated so it can be challenged, and the row names what would settle it. Against: one documented siting determination accepted across jurisdictions with no preempting standard behind it. For: a state adopting a siting standard for certified types, with placement volume tracked before and after. Alongside: Gate 0’s enumeration of which siting provisions are parameters and which are judgements — the magnitude half, still unmeasured (E-131). Until one of those exists, this beat is the argument’s furthest honest reach, and it is graded accordingly.
That is the conditional this argument was built to reach. Who gains if it holds, and who has to be paid to allow it, is the honest ledger.
Nine parties, and a column saying which figures are measured and which are blank.
| Who gains | What changes for them | How much | On what basis |
|---|---|---|---|
| The builder | Certainty before optioning land; capital turns four times as often | WITHDRAWN Aug 2026 | Rows E-070/E-071 compared site-built today with factory-built tomorrow — not like for like |
| The buyer | A price quoted before commitment, and a mortgage rather than a chattel loan | WITHDRAWN Aug 2026 | Same two rows as the builder above. Withdrawing one and keeping the other was the inconsistency — correction C-018 |
| The renter | Supply that actually responds | Rents 23% below counterfactual | Measured — Auckland, at eight years |
| The lender | A valuation record instead of an appraisal resolved at closing | Closes a gap where the loan is refused 42% of the time against 7% for site-built, at 5.61% vs 1.20% spreads | Measured — the gap (E-112), not the fix |
| The jurisdiction | Throughput and instrumentation without adding staff | 8 agencies, one consolidated answer within 20 working days | In production — Singapore; the 20 days is a service ceiling, not a measured average |
| The producer | Forecastable utilization against a committed book | — | Unquantified — nobody has tested it |
| The insurer | A priced, bounded, repeatable line where none exists | — | Unquantified — never written |
| The legislator | A reform that can be shown to have worked | — | Unquantified — no state instruments this |
| The incumbent homeowner | Nothing — and holds a share of the $48 trillion that nearby supply dilutes | The opposition, and it is rational | Structural |
Two rows are measured, one is in production abroad — and every one of those happened somewhere else. Two are withdrawn: they were arithmetic on E-070 and E-071, and the comparison underneath them was not like for like. Three are honestly blank — no aggregate figure for the economy is offered here, because nobody has produced one. That leaves nothing on this page claimed for the builder or the buyer, which is the honest state of it.
The builder gets certainty before optioning land. The buyer gets a price quoted before commitment. The renter gets supply that actually responds. The lender gets a valuation record instead of an appraisal resolved at closing; the insurer gets a priced, bounded line where none exists; the producer gets forecastable utilization. The jurisdiction gets throughput without adding staff. And the incumbent homeowner — the row that decides whether any of it happens — gets nothing on this list, which is the problem.
InvestorsEight aligned parties and one structural opponent is an unusually good ratio for housing, and it is why segment selection is the first move rather than an afterthought. Note also which rows are blank: the producer, the insurer and the legislator gain in ways nobody has yet put a number on.
PolicymakersThe last row is the political problem, and argument does not solve it — choosing ground where the incumbent is also the builder does. Arizona is the test: casita bills passed, starter-homes died twice.
ResearchNote how many rows read unquantified. That is the honest state of this, not a gap in the table — and the opponent's row is not ignorance. Any version of this argument that treats opposition as a communications problem has misread the structure it is describing.
Those are the stakes, blanks shown. Now the arguments against.
Stated in their strongest form. Three are answered. Eight are not — including the three with the best evidence behind them.
| The objection, in its strongest form | How it ends | What would settle it |
|---|---|---|
| Savings capitalize into land. You are optimizing a rounding error. | Still open | Auckland answered it for renters, not for buyers of a new type |
| Codes will not compile — discretion is the point, not an accident. | Test first | Gate 0. Two weeks, no capital |
| BIM code-checking was tried in America and went nowhere. | Unresolved | Now modelled, and it moved the buyer: acceptance is won at a state legislature, not sold to a department. Untested — E-108 |
| This is permitting software — a crowded, low-margin category. | Answered | Legal weight, not customer, is the distinction |
| Type certification means every house looks the same. | Answered | Type approval governs the envelope, not the appearance — and Japan varies the product per buyer inside it, at 13% of starts. E-109 |
| The real mechanism is firm size, and your fix may make it worse. | Conceded | The best causal evidence on the page, and it cuts against us |
| Every precedent you cite was political, not technological. | Standing risk | Correct on the premise; the two are complements, not substitutes |
| Trade labor is the constraint and no compiler fixes it. | Answered | Partly idle capacity; the residual is real and not claimed solved |
| A jurisdiction can just buy the automation itself and skip all of this. | Unresolved | Honolulu requires an automated pre-application screen from September 2026 — no outside determination, no professional of record, no carrier, no portability, and the department still decides (E-117). If that path captures most of the available time, this architecture answers a question the market is closing by a simpler route. Nobody knows: the vendor’s halving figure is the vendor’s (E-118, Untested), the city publishes nothing, and what the screening market was worth was never measured either (E-060) |
| The reviewer is paid by the party it reviews — and aviation ran that experiment. | Conceded | 24 CFR §3282.351(e) lets a primary inspection agency contract directly with the manufacturers whose designs it approves. That is the structure of FAA delegation, whose inspector general found “management and oversight weaknesses limit FAA’s ability to assess and mitigate risks” with the Boeing ODA, and unit members reporting pressure to affirm compliance without time to review — E-115. HUD writes real controls: four monitoring visits a year, disqualification, a third-party petition right, excessive fees as their own ground for removal — E-116. Whether any of it has ever bound is unmeasured, and a written control is not a demonstrated one. What would settle it: a count of DAPIA disqualifications, fee removals and third-party petitions since 1976, with outcomes — a records request to HUD, not a research programme (E-119) |
| Prefab typically costs no less than conventional construction, so the factory-cost advantage is illusory. | Conceded | E-098. A like-for-like price per square foot, which this project cannot currently produce |
Eleven objections. Three answered, eight ending in a concession — and the two with the best evidence behind them are both against this argument. The prefab-cost objection has an available answer that is deliberately not claimed: the HUD-code home is a different product from modular built to local codes — federally preempted under 42 U.S.C. §5403(d), with no site-code compliance to satisfy — and it averaged $137,500 in March 2026, against a median existing home several times that. That is not a like-for-like comparison, because it sets a whole different product and a different bundle of land and finish against each other, so it is not offered as one. Until E-098 carries a real per-square-foot comparison at matched finish and region, this objection stands.
Savings capitalize into land. Codes will not compile, because discretion is the point. BIM code-checking was tried here and went nowhere. This is just permitting software. Type certification means every house looks the same. The real mechanism is firm size. Every precedent cited was political, not technological. Trade labor is the constraint. And, added in August 2026: prefabrication typically costs no less than conventional construction, so the factory-cost advantage is illusory. And, days later, the tenth: the reviewer is paid by the party it reviews, which is the structure of the strongest American precedent this argument has. And the eleventh, from the same week's research: a jurisdiction can simply buy the automation itself. Three of those are answered. The other eight end in a concession.
InvestorsEight of the eleven end in a concession. Three are worth reading twice. The firm-size objection has the best causal evidence on the page, is conceded outright, and is in tension with the small-builder claim — both cannot be the affordability mechanism, and that reconciliation is owed before any raise. The prefab-cost objection is conceded because the available answer — that a HUD-code home is a different product from modular built to local codes — needs a like-for-like price comparison this project cannot yet produce. The third came from our own reading: the delegated-review precedent this argument leans on has the reviewer paid by the manufacturer it reviews, and aviation's version of that was found inadequately overseen by its own inspector general. HUD writes real controls; nobody has measured whether they bind.
PolicymakersOn the monotony objection, the answer is not a promise: Japan builds roughly an eighth of its housing starts in factories and varies the product per buyer inside a standardized system, which is what type approval permits everywhere it exists. Standardize what is checked, not what is seen. The land-capitalization objection is the one a housing economist will raise first, and it is now partially answered by measurement rather than argument: Auckland's rents came in 23% below counterfactual at eight years. Capitalization is real, but it is not total.
ResearchEach objection is stated at full strength and answered or conceded on the record, with the thing that would settle it named. The count has moved four times — nine and three, which was wrong; eight and five; ten and seven when 24 CFR §3282.351(e) turned out to say that this project's best American precedent has the reviewer paid by the party it reviews; now eleven and eight, since Honolulu began requiring automated pre-application review that needs none of this architecture. Every move has been upward, and every addition has been a concession.
Eight concessions, and most of them stand. Not one of them has been started on.
This is the whole plan, and every status on it currently reads: not yet.
Each link needs the one above it. Two of the five turn out to be precedented where a single standard governs — which is an argument for preemption, not evidence that a determination travels between jurisdictions that keep their own rules. Nothing here has been demonstrated across rulesets. The first link is two weeks of work.
Codes have to prove compilable. A jurisdiction has to accept a Lintel determination in place of first-pass review. A carrier has to insure that determination. A lender has to value a certified type rather than a comparable sale. A producer has to build against a committed book. Each link needs the one above it, which is why the order is not a preference — and every one of them is currently unproven. The order is itself falsifiable, and the conditions are named: if Arizona's granted right produced throughput unaided, information demotes to a convenience; if builders name utility capacity rather than approval as the binding gate, the middle link is aimed wrongly; if binding provisions prove predominantly discretionary, the chain collapses to politics plus capital — the manufactured-housing path with modern finance.
InvestorsThis is a dependency chain, not a schedule, and it is deliberately not a schedule — dates would be invented and the chain is what actually governs. Note where the value concentrates: links two and three are the wedge, link four is the endgame.
PolicymakersTwo of these five are things a legislature can act on directly. A state that wanted this could authorize acceptance of an insured determination — link two — and separately require its jurisdictions to publish permit data, which is Gate 1 and needs no product from anyone.
ResearchAmerican law already reaches part of link two: several states preempt local authority over approved industrialized buildings, and interstate reciprocity recognizes streamlined certification across state lines. None covers the site-plan scope, which is the gap.
The first link is two weeks of work. And the rows it would fill are the ones this argument leans on hardest.
The claims the argument leans on hardest are among the least evidenced. That is the shape of the opportunity.
And separately, both narrower than this site used to say: a jurisdiction accepting an outside determination is done where one standard governs — Florida since 2006, a private HUD-accepted agency’s design approval good in every state since 1976 — and a carrier already prices that risk, because Florida sets the reviewer’s cover in statute. What is undemonstrated is either of them across jurisdictions that keep their own rules.
E-039 — the share of binding provisions that are objective rather than discretionary. Never measured, anywhere. E-107 — the placement gap: the share of residential parcels where a dwelling is permitted but a home someone can actually order is not placeable. Never measured, and now the target of Gate 0. E-037 — Arizona permit volume since preemption. Not published. E-038 — the full idea-to-sale cycle. Never measured end to end. E-098 — a like-for-like price per square foot, factory against site-built, which is what the strongest objection on this site turns on. And separately: whether a jurisdiction will accept an authoritative determination, and whether a carrier will insure it. Neither has been demonstrated.
InvestorsNo gate has been cleared yet. This is a working thesis, pre-evidence-collection, and it is being presented as one.
PolicymakersSeveral of these are things a state could measure for itself, cheaply, and does not. That absence is the most portable finding in this talk.
ResearchGaps are register rows too, recorded so they cannot be quietly forgotten. They are the research contribution, not a to-do list. The count went from eleven to fifteen in August 2026, and to thirty-two since — the correct direction for a project doing research, and the opposite of the direction a pitch moves in.
Both gaps are cheap to close. What would settle the whole thing is cheaper still.
All three precede any capital commitment. All three are publishable whichever way they come out.
The Florida read · days, not weeks · public records · fills E-108
Florida has allowed a private provider to review plans in lieu of the building department since 2006 — insured, clock-bound, automated review permitted since 2025. What share of permits use it has never been read.
EITHER WAY — high uptake means the trade works and portability is the open question. Near-zero after nineteen lawful years is the bigger obstacle.
Gate 0 · one county, three jurisdictions · no capital · fills E-107
Rewritten August 2026. On the National Zoning Atlas method, record each district’s manufactured-housing status and the standards that bite, then test three purchasable models against every parcel: one counts only if an orderable home clears every standard.
KILL CRITERION — if binding provisions are predominantly discretionary, the exclusion is deliberate and legal reform, not information, is the lever.
Gate 1 · ~4 weeks · no capital · fills E-037
Records requests across Arizona for accessory-dwelling permit volumes, 2023–2026, against comparable metros. Ten to fifteen builder interviews mapping which gate actually binds.
EITHER WAY — that these figures are unpublished is itself a finding: a state reformed and built no instrumentation to see whether it worked.
The Florida read, days: Florida has allowed a private provider to perform plans review in lieu of the building department since 2006, insured and clock-bound. Public permit records say what share of permits actually use it. Gate 0, two weeks: on the National Zoning Atlas method, record each residential district's manufactured-housing status and the standards that bite, then test three purchasable models against every parcel — one counts only if a home someone can order clears every standard. Gate 1, four weeks: records requests across Arizona for accessory-dwelling permit volumes 2023–2026, plus ten to fifteen builder interviews mapping which gate actually binds.
InvestorsRun the Florida read first. It is days of records work, and it tests what the gates do not reach: not whether rules can be computed, but whether anyone accepts a computation they did not perform. Near-zero uptake after nineteen lawful years would be the most important negative result available here.
PolicymakersTwo of the three are records requests. Any state could run Gate 1 on its own reforms, and Florida's own data would say whether a mechanism it enacted in 2006 is used at all. Both findings belong in the public record regardless of who collects them.
ResearchEven a negative result is publishable and novel — nobody has measured the objective/discretionary split of American zoning. A paper reporting it would be cited regardless of which way it comes out.
Run any one of them and this argument changes. That is the point.
The ask is not money. It is the test.
Falsification statement — filed before evidence collection, August 2026
This thesis fails if (a) binding provisions prove predominantly discretionary and therefore non-compilable; or (b) encoding cost per jurisdiction does not fall below the threshold implied by the addressable market; or (c) builders predominantly identify utility capacity or fire-district sign-off, rather than approval-side gates, as the critical path on recent projects.
It is materially weakened if the firm-size mechanism dominates — in which case type certification addresses a problem adjacent to the real one.
Clause (b) has now been tested, and it failed in a way the statement did not anticipate. Encoding cost was never the binding constraint: Singapore's engine was handed to four other jurisdictions who inherited it for free, and none reached production. The claim built on clause (b) is withdrawn from this site — see the revisions page and correction C-013. Clauses (a) and (c) remain untested. This is what the statement was for.
This thesis fails if binding provisions prove predominantly discretionary and therefore non-compilable; or if encoding cost per jurisdiction does not fall below the threshold implied by the addressable market; or if builders predominantly identify utility capacity or fire-district sign-off, rather than approval-side gates, as the critical path on recent projects. It is materially weakened if the firm-size mechanism dominates. Every one of those is cheap to test and none has been tested. That is the current state of this project, stated plainly.
InvestorsFiled before evidence collection, in August 2026, in public. Everything here is graded, sourced and revisable — and the first finding gets published under my own name whichever way it comes out.
PolicymakersThe offer is instrumentation. A reform that nobody measures cannot be defended when it is challenged, and right now none of them are being measured.
ResearchStated before the evidence rather than after it, which is the only time a falsification statement is worth anything.
And if only three things survive this, make it these.
If nothing else survives this, these should — and each one fits in a sentence.
Thirty-two months to deliver an American home. Eight of them build it. The other twenty-four are information, waiting and risk — and that is the part nobody has attacked.
Politics decides whether you may enter, information whether anything moves, capital whether it survives; four serious attempts each turned one lock and were killed by another. Everything in this record that travels rides on a preempting standard, and siting has none. If that pattern holds, the missing instrument is a preempting siting standard — an inference, not a finding, graded Untested (E-229), with fifty legislatures on a dated clock to 11 July 2027 (E-197).
A 25% margin over thirty-two months annualises to about 9% — below the hurdle of almost every risk-capital pool, which is why housing is funded by patient money or not at all. The same margin over seven months is about 47%, and a 10% margin over seven months still returns about 18%. Shorten the clock and the risk-reward ratio admits capital that cannot participate now. Modelled, never run — rows E-070 and E-071, both Untested.
One: it takes thirty-two months to deliver an American home and only eight of them build it — the rest is information, waiting and risk, and that is the part nobody has ever attacked. Two: three things block it, in a fixed order — politics decides whether you may enter, information decides whether anything moves, capital decides whether it survives — and four serious attempts each turned one lock and were killed by another. What the record adds is a pattern: everything in it that travels — a design approval good in fifty states since 1976, a vehicle certificate good in fifty — rides on a preempting standard, and nothing about where a home may go travels at all, because siting has none. If that pattern holds, the missing instrument is a preempting siting standard. That sentence is an inference, not a finding — graded Untested in this project’s register (E-229), with what would settle it named on the row — and fifty legislatures are already on a dated clock, most to 11 July 2027, that reopens exactly the statutes it concerns (E-197). Three: an optimized timeline is a new asset class. A 25% margin over thirty-two months annualises to about 9% — below the hurdle of almost every risk-capital pool, which is why housing is funded by patient money or not at all. The same margin over seven months is about 47%, and a 10% margin over seven months still returns about 18%: better than today's return at less than half the margin. Shorten the clock and the risk-reward ratio changes enough to admit capital that cannot participate now — which is the mechanism by which supply moves. It is modelled and never run (E-070, E-071, both Untested), and it is the proposition this project exists to test rather than a result it reports.
InvestorsRead the third one as due diligence rather than as candour. A thesis that publishes its own refutation is telling you what its evidence is actually worth, and what it is worth here is: no gate cleared, thirty-two rows empty, and one central claim withdrawn.
PolicymakersTwo of the thirty-two empty rows are things a state could measure on its own reforms, cheaply, and none currently does. That absence is the most portable finding here.
ResearchEvery number behind those three sentences is graded and sourced, thirty-two of the supporting rows are deliberately empty, and the falsification statement was filed before any evidence was collected. One of its three clauses has now been tested and the thesis lost it.