Housing as a Product — a systems thesis on American housing production
The other twenty-four are paper. America’s building rules were never written in a form a computer can check, so every house is approved by hand — one project at a time, in each of twenty thousand jurisdictions that write their own.
Of those twenty-four, eighteen carry no physical work at all; the remaining six are site and sale. This thesis claims one block of the eighteen — the five months of plan review, where work that is already certified is verified a second time. It does not claim the eleven months of entitlement. Those provisions are discretionary, they stay with the jurisdiction, and the design refers them rather than clearing them. How much a compiled determination would actually remove has never been measured — that is register row E-060, and it is empty.
The fix argued here is one transaction: a computer checks a house design against one city’s rules, on one lot; a licensed professional signs the result; an insurance company pays if it is wrong; and the city accepts that instead of its own first review. No city accepts one today, and no insurer writes one — that is the untested core of this thesis.
Download the full report · PDF Corrections since it was closed are on errata.
Each square below is one month on the critical path, from site identification to closing. The solid squares are not the problem. They are the product. Everything else is overhead.
Scroll the strip sideways to see all 32 months →
Today · 32 months
Withdrawn · August 2026
The instrument reads back— hover a square, or focus the strip and use ← →
Sources · US Census Survey of Construction: single-family authorization to start plus start to completion, 8.8 months for 2025 (register row E-002 — and note these are two averages on different cohorts, not one figure). The eight months of building is the 2021–2025 mean of 7.84, not any single year (E-139); the annual series runs 6.8 to 8.6 and a headline pinned to the latest vintage would move most years without anything changing on site. Entitlement commonly 6–18 months, 18–36 acquisition-to-entitled for complex projects; California subdivision routinely 3–7+ years; Gabriel & Kung (Journal of Urban Economics): requiring a full environmental impact report added 484 days on average to approval time in Los Angeles. Counter-evidence · the only two projects this register measures end to end run the other way: Florence Towne Apartments, 1,725 days — 465 approval and 1,260 construction (E-067) — and the Los Angeles multifamily population it sits in, 1,413 days, 652 approval and 863 construction (E-068). Both are multifamily, both are one city, and both start at the entitlement filing, so neither counts feasibility, site or sale. They measure a different window from this strip, and on the window they do measure they are evidence against the ordering asserted above rather than for it. Caveat · today's critical path is assembled from separate sources never designed to be added together — no dataset tracks a single project from site identification to closing (register row E-038). The compiled row is an estimate, not a measurement. Abroad · that is not an American failing. No country in a 38-country compilation publishes approval, waiting and construction as one regular series, and where approval does dominate it sits beside places where it plainly does not — what the world publishes. At home · nor is it only abroad. The two states that compel a jurisdiction to report how long permitting takes publish no permit duration between them; joining California's own two tables yields one for 190 of 529 jurisdictions, and the rest file something no duration can be computed from — what two states actually publish.
One word, defined
A rule is compiled when a computer can check it: the way a tax form computes a number, not the way an essay is graded. That is the only piece of vocabulary this site asks you to carry, and it is why the argument is made in the language of software, about buildings.
One buildingA recent real world example
Fifty-one affordable homes at 410 E. Florence Ave., South Los Angeles. The thirty-two months above is an average assembled from separate sources. This is one real building, with dates.
1,725 days. Four years and nine months for fifty-one homes — and 465 of those days passed before anyone was permitted to start building. Not an outlier: across every multifamily project permitted in Los Angeles between 2010 and 2022, average total development time was 1,413 days. Both figures are register rows E-067 and E-068, from a peer-reviewed study of the city's own permit records — this project is a real one, cited rather than illustrated.
If this is the problem, the first test of the fix is cheap. If you want the Gate 0 number when it exists — the first measurement of how much residential land a home you can actually buy is legally barred from — send an email and you will get it, including if it comes out against this thesis. No date is promised here, because none has been earned: nothing on this site is scheduled that depends on someone else saying yes. What Gate 0 is ↓ · the decision rule, committed before the data →
The size of itWhat this is a problem with
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.
What it issuesOne artifact, and what it does not claim
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.
Untested Thirty-two of two hundred and thirty-one rows are unmeasured. Three are load-bearing. They are E-037, E-039 and E-051 — whether approval is what actually binds a builder, what share of American zoning is machine-checkable, and what compiling one jurisdiction now costs. Each is cheap to settle, none has been settled, and what would settle each is stated below.
One measured factWhat already happened, before any proposal
American factories shipped 373,143 manufactured homes in 1998. By 2004 they shipped 130,748 — a 65% fall, before the financial crisis, while site-built housing was having its biggest boom in decades. The floor came later: 49,717 in 2009. This is the cheapest housing the country knows how to build, collapsing against the cycle.
The HUD Code federalised how these homes are built in 1976 and said nothing about how they are owned. States title them through motor vehicle agencies, so the cheapest new house in America is bought like a car: chattel paper, not a mortgage. Manufactured-home purchase applications are denied 42% of the time against 7% for site-built, and the gap survives controlling for credit score — in the super-prime band above 720, chattel approval is 63% against 95% site-built.
NoteCensus Manufactured Housing Survey, US total row of the annual shipments table. Shipments have since recovered above 100,000 — 112,882 in 2022, 102,738 in 2025 — and remain under a third of the peak; the collapse is real and it is not permanent. 1998, 2004 and 2009 all sit before the 2014 methodology change, so these comparisons are internally clean. Rows E-012, E-112, E-113 — measured history, not a projection. This site claims no outcome figure for anything it proposes.
The short versionThree things, if you read nothing else
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.
ThenEverything below, and everything on this site, is the long form of those three sentences — every number graded, sourced, and open to being wrong.
The argumentFive sentences
One note on emphasis. Sentence two says the cause is that the rules are prose. That is the instrument, not the destination — compilation is the wedge; type-based valuation is what the thesis is actually for. Three locks bind here, in a fixed order, and information is the middle one. See the sequence →
What is honestly unproven — read this before the rest
American codes are compilably objective. The share of binding provisions that are objective rather than discretionary has never been measured, anywhere. The single most load-bearing claim on this site, and register row E-039 is empty.
A jurisdiction will accept an authoritative determination, and a carrier will insure it. The entire thesis rides on acceptance and insurability, and neither has been demonstrated.
Arizona's granted right produced throughput. Statewide by-right casita rights took effect in 2026; permit volumes since are published nowhere. Register row E-037 is empty.
No gate has been cleared. This is a working thesis, pre-evidence-collection — every claim on this site is graded and sourced in the evidence register, and the three gaps above are the research agenda, not a footnote.
And what this would do for someone renting. The mechanism is indirect and worth stating plainly: if approval time falls, the cost of carrying land and capital through it falls, and more of what gets built is the small infill a single lot can take — an accessory dwelling, a duplex on a corner. More homes in a place, sooner, is what would eventually reach a renter. No figure is claimed for that, and the absence is deliberate. The only resident-outcome number on this page belongs to Auckland's upzoning, which is a different intervention, and the row that once put a figure on the buyer's side was withdrawn in August 2026 and is recorded as withdrawn. Nobody has measured what a compiled determination does to rents, so this site does not say.
What would settle itTwo tests · about six weeks · no capital
The three Untested claims above are not a caveat at the end of a long argument. They are the work, and the first two are reachable in six weeks by one person with no money and no permission.
Where can a home you can actually buy legally go? One county, three jurisdictions, no capital. Rewritten and narrowed in August 2026. For every residential district record manufactured-housing status and the standards that bite — minimum width, floor area, roof pitch, exterior materials, foundation, unit age — using the National Zoning Atlas method so the result is comparable to the atlases that already exist. Join parcels to districts, then test three real, purchasable models against every parcel. A parcel counts only if a home someone can order clears every standard. Fills register rows E-107 and E-039. Kill criterion: if binding placement provisions are predominantly discretionary, the exclusion is deliberate and legal reform rather than information is the lever. What each outcome will be taken to mean →
Does a granted right produce throughput? About four weeks, no capital. Records requests across Arizona municipalities and counties for accessory-dwelling permit volumes 2023–2026, synthetic control against comparable metros, plus ten to fifteen builder interviews mapping which gate actually binds. Fills register row E-037.
Neither requires capital, permission or partners, and both results get published whichever way they come out — that these figures are unpublished is itself a finding. Neither has been run. The full method →
Four doorsWhere to go next — by what you came here to do
01 · 17 or 74 min
The whole case as one continuous narrative in five movements — the gap, the 1976 standard that stopped at construction, the quarantine that followed, what four more attempts prove, and an ending that concedes what it has not proved. Read it brief — one line per beat, ten minutes — or in full, through the investor, policymaker or research lens.
→ /story
02 · 11 min
Lintel on its own page — a conformance authority: a compiled determination, signed by a professional of record, insured, and accepted in place of first-pass review. With the sequence that forces its shape — three locks, politics then information then capital, binding in a strict order that every serious attempt broke in a way you can name. What is already settled in law beneath it, and the two things nobody has ever demonstrated.
→ /solution
03 · PDF
HAP-02, thirteen sheets with works cited — the citable snapshot, alongside every other document this project has published. The argument itself is maintained in the story and the instrument; the paper is what goes in a footnote.
→ /documents
04 · browse
Two hundred and thirty-one claims, each with source, grade and review date. Sortable, filterable, and honest about the thirty-two rows nobody has measured.
→ /evidence
Research — everything this project has measured, in one place: what two states publish, which clock a permitting figure is running, and what 38 countries do not publish; The research log — every position this project has held, tested and changed, in order, with what each change cost; what changed in August 2026, when the claim at the centre of this argument was refuted by its own research and removed; and what was already known, which says who published this before we did. Also — the certification window, the ledger of what the July 2026 Act obliges each state to file by 11 July 2027 and what has actually been published · the same fifty-four beats as a keyboard-driven deck, with presenter notes and the same three lenses · the corrections log, everything later found wrong or weaker than stated, kept on the page after it is fixed · every document, with revision and date.