Housing: why doesn't legalizing it build it?

The number that isn't: the housing shortage is an assumption in a trench coat

Nearly every mainstream shortage estimate belongs to one method family: pick a "target" vacancy rate, subtract reality, call the difference the shortage. Running that model on committed Census data, one percentage point of assumed target vacancy is worth about 1.59 million units — and feeding each publisher's own stated target into the bare model reproduces Freddie Mac's famous 3.7 million to within 116,000 units from the assumption alone. The published range (5% to 12% targets) spans more units than the entire spread of published estimates. Two federal surveys agree on how many housing units exist to within 0.06% and disagree on how many are empty by over a million.

Legal on paper, unbuildable in fact: the design detail that decides it

Same state, same era, two designs. California's ADU laws — prescriptive, ministerial, envelope-adding — took production from about 540 permits in 2016 to over 25,000 a year, a fifth of the state's housing output. SB 9, which legalized more units per lot but left local standards and discretion intact, produced 38 approved units in Los Angeles and zero lot splits; eight of thirteen surveyed cities approved none. Austin repeated the experiment independently: its HOME reform legalized three units per lot while leaving the buildable envelope at 0.4 FAR, and the three-to-four-unit permit class declined after the reform legalizing it — 51 units in 2023, 8 in 2024, 18 in 2025, still roughly a third of the pre-reform year two years on. Querying the same database by unit count, the HOME three-unit path produced about 15 permits in two years, against 233 permits at two units. Both come from the city's own permit database, not press coverage. Two limits travel with them: that Austin books a HOME three-plex into the three-and-four-family class is confirmed by that unit-count query, not against Austin’s written permit-classification guidance; and the same series carries a classification break — the two-family class rises 156 → 532 while the secondary-apartment class collapses 248 → 10 over the same window — so the small-scale totals cannot be read as pure new production.

Every reform that added buildable envelope produced homes. Every reform that raised the unit count while holding the envelope produced almost none — two cities, two states, administrative data, mechanism documented in each city's own ordinance record.

Who actually blocks housing — and it isn't about their home values

The standard theory says homeowners block housing to defend their asset values. The evidence runs the other way: upzoning tends to raise the value of the upzoned lots, and homeowners oppose it anyway — which means the binding motive is something else (neighborhood character, parking, control), and instruments built to compensate asset losses are aimed at a mechanism that isn't operating. What actually killed the recent wave of state bills wasn't the tenant-vs-developer fight the press covers: New York's Housing Compact and Colorado's 2023 preemption bill both died to suburban home-rule revolts inside the governing party. And the swing vote with the best-documented arc is construction labor — California's Building Trades killed the 2018 and 2020 upzoning bills, then endorsed or went neutral on the 2023 and 2025 ones after enforceable labor standards were written in. The bills passed with the tenant coalition still opposed.

The workers nobody's plan supplies

Closing any defensible shortage band requires residential-construction workforce growth between 2.6 and 63 times the sector’s recent trend, depending on the band and the ramp — 3.8 to 31 times on a ten-year ramp. The channel that has quietly absorbed the gap is immigration — a record 26.3% of the construction workforce is foreign-born, roughly one in three in the trades, with the undocumented share modeled at 15–23% — a band assembled from three organizations estimating three different populations (all construction workers, laborers only, Texas only), not three measurements of one quantity. That channel is exposed to near-term enforcement policy, not slow demographics. Meanwhile the textbook signature of a binding labor shortage — a construction-specific wage premium — is absent: construction wages grew fractionally slower than the private-sector average over 2020–2026. Hard costs, it turns out, are not the lever in constrained metros at all; land and process are.

The leaky plumbing: vouchers landlords refuse, a subsidy nobody can price

Federal rental assistance is rationed, not an entitlement: three in four income-eligible households receive nothing. The flagship demand-side instrument leaks where it matters most — landlords refuse vouchers at 76–78% rates in the two study markets with no source-of-income protection, and at 67% in a third with partial protection, and refusal is highest in low-poverty neighborhoods, exactly where the voucher's mobility purpose points. That leak is fixable, and this filing missed the evidence: a randomized trial in Seattle and King County paired vouchers with search help and direct landlord engagement and moved families into high-opportunity neighborhoods at 53% against 15% for the control group, at $2,670 a household — while a straight increase in what the voucher pays in expensive neighborhoods moved almost nobody. On the supply side, the flagship program is a measurement black box: no federal agency is assigned to maintain and analyze LIHTC cost data, the widely-assumed cost penalty versus market construction has never been tested nationally (and state studies point the other way), and the roughly 17-cent gap between a credit dollar and the equity it buys cannot be decomposed into time value, investor return, and fees from any public source.

Insurance eats the subsidy

Multifamily property insurance premiums roughly doubled between 2021 and 2024 — more than six times the rise in CPI over the same period, and up ~45% in 2023–24 alone. The asymmetry is the finding: a market-rate owner passes an insurance shock through to rents; a rent-restricted owner cannot — rent restriction is the program's core mechanism — so the same shock that market housing passes along, affordable housing absorbs out of net operating income until maintenance defers or the asset is lost. Every architecture in the standard reform debate is a capital-side or demand-side instrument; none touches the operating side. That diagnosis comes straight from the workstream evidence. The candidate fixes (insurance pooling, backstops, parametric reinsurance) do not — they are this desk's suggestions, labeled as such.

The cap that isn’t: 278,345 public-housing units already qualify for federal funding

The Faircloth limit is usually described as a ban on new public housing. It is actually a condition on two federal funding streams: a housing authority may not spend Capital or Operating Fund money building above the unit count it held on October 1, 1999 — and the statute’s own subparagraph (B) lets it build above that line anyway, just without extra formula money. Because the country demolished so much public housing since (HOPE VI alone removed 93,295 units), that funding line now sits 278,345 units above the standing stock. This project pulled HUD's own published Faircloth List and summed it directly — twice, independently, with a committed pipeline whose identity check caught a 13-unit inconsistency inside HUD's own file. A quarter of the headroom sits in just five housing authorities. The binding constraint on public housing is money and capacity, not legal authority — and the fall from a 1.4-million-unit peak to 878,000 units standing today is a completed durability test any revival plan has to answer.

What other places prove — and what they can't promise

Tokyo out-built all of California — 124,810 starts against 111,760 permits in comparable years, roughly 3.2× per capita, and the standard caveat about disposable Japanese houses didn't survive root-tracing (the two housing stocks turn over at comparable rates, making Tokyo's advantage larger, not smaller). Auckland's celebrated upzoning result is contested — every favorable estimate runs through one author, with three confounders inside the window. Vienna's social housing is real and enviable, but its enabling condition — a land bank acquired over a century starting when rent control had crushed land values — is non-portable, and Vienna's own peer-reviewed literature shows land scarcity now constraining the model at home. Singapore’s system rests on compulsory land acquisition at below-market prices — and it is the below-market pricing, not the compulsion, that the Fifth Amendment’s just-compensation requirement forecloses. Sweden's queue is the cautionary tale done precisely: nine years average in Stockholm, 15–25 in prime districts, with a documented black market. And the American postwar miracle was a peak-year artifact — 1950's ~1.95 million starts, not a decade-long average, on financing built atop explicitly discriminatory underwriting no one should copy.

The scorecard, twice-checked and then steelmanned: the lead is one cell wide

Nine seeded architectures, four objectives, six weightings — and two full adversarial checks. A red team caught the first draft scoring its favorites as idealized composites; then a second scorer, blinded by instruction rather than by structure, caught the correction itself still enforcing the no-evidence-no-score rule in only one direction — nine cells sat below neutral on mechanism hunches, three cells missed evidence that was sitting in the record. Reconciled, the honest result: this record produced above-neutral evidence for exactly one instrument — envelope-prescriptive legal reform — cautions industrialized construction, whose scaling thesis the record refutes, carries one genuinely mixed verdict (Faircloth social housing: real legal authority, adverse operating history), and is honestly agnostic on the rest. Most of what looked like ranking was evidence-density. Housing-First stays unscored. Both scorers, working blind of each other, independently agreed the record then held nothing to score it on; a late evidence pass (§12, 2026-08-06) has since supplied one, and the row moves only after a fresh matrix pass.

A third check — the Phase 2 steelman, 2026-08-10 — found that "leads every weighting" is thinner than it reads, and withdrew one caution. Twenty-three of this board's thirty-two cells are neutral, and envelope-prescriptive preemption holds the only cell above neutral anywhere on it — so its lead follows from the shape of the matrix rather than from running the weightings, and it is worth a quarter of one point. That cell is also the one our two adversarial checks split on: the red team argued it down to 3, the blind second scorer marked it 5, and it was reconciled to 4. Read at the red team's 3, envelope-prescriptive preemption leads none of the 1,771 weightings we swept and ties the land-value tax, the row this filing itself calls an ignorance artifact. The arithmetic is now committed and reproducible. And the caution against the bare voucher does not survive: HUD's own randomized experiment across 2,282 families in 12 cities cut homelessness from 38% to 17% using an ordinary housing voucher, and a second trial fixed the voucher's mobility failure for $2,670 a household. Neither study appears anywhere in this filing's record. The caution is narrowed to what it was always about — a voucher builds no housing — and withdrawn as a verdict on the instrument.