What a delivery time needs, and what anybody publishes
Every published figure for how long housing takes is assembled from parts that were never meant to be added. Not because the data is hard to gather — a public body already holds the application date and the decision date for the same building — but because almost nobody publishes the subtraction. This is what the gap looks like measured, in the two American states that compel the reporting and across the 38 countries two studies could compile, and what would close it.
This is not a ranking of states, and there are no other states in it. The selection rule is the mandate — the states that already compel the reporting — and exactly two have one, so this is the whole population rather than a sample. The two are never added, averaged or placed in one column: they measure different intervals, on different populations, with different rules about when the clock starts. Both are permit durations and neither is the other.
The gapThree phases, and nobody keeps all three
Delivering a home has an approval phase, a waiting phase and a construction phase. To say how long delivery takes you need all three, measured on one population, by one publisher. No authority anywhere publishes that.
Two studies commissioned for this project compiled what is published across 38 countries. Of 183 country-by-segment records, 13 carry all three phases at once — and every one of those splices two publishers, splices two years, or puts a statutory deadline where a measured approval time should be. Coverage of the single easiest phase is thinner than the scope suggests: 60 of 183 records carry a construction figure, and 26 of 38 countries have one at all.
So every published comparison of how long housing takes — including any this project could make — is assembled from parts that were never meant to be added. That is the gap. The rest of this page is what happens when you try to close it from the records that do exist, in the two American states that compel a local government to report its own timing.
RegisterE-140supportedE-141confirmedE-134confirmedcheck these rows →
CaliforniaThe dates are published. The duration is not.
Every city and county in California files a Housing Element Annual Progress Report. Table A carries the date each housing application was submitted. Table A2 carries the dates it was entitled, permitted and occupied. The state publishes both and joins neither.
Joined on jurisdiction and project identifier, 260,811 of 330,855 distinct projects match — a match rate of 78.8%. That was the number expected to be the constraint, and it was not: the Terner Center matched roughly 29,000 of about 60,000 doing this for ADUs, and two keys that fail on different projects do better than one. 133,088 of the matched projects carry a usable application-to-issuance duration once the cohort is restricted to applications with three years of follow-up.
The constraint turned out to be what the matched dates mean. 13.7% of matched projects carry an identical application and issuance date, and 8,992 more carry a permit date that precedes the application. A nought-day elapsed time is a jurisdiction entering one date in two fields, not a permit issued the day it was asked for.
RegisterE-184confirmedE-186confirmedcheck these rows →
The screenThe defect is a property of the jurisdiction
The contamination is concentrated, not spread. Los Angeles sits at 1.4% across 29,938 projects. Bakersfield sits at 81.5%. That difference is the finding: data quality is a property of the jurisdiction, so a statewide median computed over every filing is contaminated in a way a per-jurisdiction one is not.
A median is therefore published for a jurisdiction only where that jurisdiction's own record survives two tests — at least 30 matched projects, and at most 5% of them carrying an identical date pair. 190 of 529 jurisdictions clear it. 339 do not: 234 have fewer than 30 matched projects and 105 fail on the dates themselves.
Both thresholds are judgment, and both are published as columns in the extract beside this page, so a reader who prefers different ones can re-run the screen rather than take this on trust. A jurisdiction that files the form and cannot produce a duration from it is recorded as a finding, not dropped as a gap.
RegisterE-185confirmedcheck these rows →
CaliforniaApplication to building permit, calendar days
Medians of jurisdiction medians, in calendar days, from application to issuance of a building permit. Entitlement is broken out and never folded in.
Multifamily, 5 units or moreOnly 10 jurisdictions clear the screen for this class, so it is the weakest figure here and the range behind it is wide.
440 days
Accessory dwelling unitAcross 138 jurisdictions — the best-covered class, because ADUs are what California jurisdictions process most of.
196 days
Single-family detachedAcross 82 jurisdictions that clear the screen.
130 days
Application to entitlement is 102 days, across 48 jurisdictions. It is deliberately not drawn on the axis above: entitlement is a different stage, not a fourth product class, and putting it on one scale with the three would invite exactly the addition the measurement spec forbids.
In the largest jurisdictions: Los Angeles 189 days overall and 633 for multifamily; Los Angeles County 417 and 891; San Diego 236; San Francisco 484. Los Angeles County at 891 days for multifamily is two and a half years from application to a building permit, before anything is built.
A project that applied and has not been permitted cannot appear in a join at all, so every figure here describes projects that finished — and the slow ones from recent years have not. The cohort is cut to applications on or before the end of 2022 to give three years of follow-up, which reduces that bias without removing it. A median of jurisdiction medians also weights a jurisdiction with 30 projects the same as one with 29,938. Neither is corrected for, because correcting either means assuming something about the projects that are missing.
RegisterE-187supportedE-188supportedE-155confirmedcheck these rows →
WashingtonThe one thing almost nobody else measures
Washington requires its seven most populous counties and the 44 cities over 20,000 within them — 51 jurisdictions — to report permit timeliness to Commerce. It then does something almost no authority anywhere does: it publishes the elapsed wait and the agency's own clock, with pauses removed, side by side.
The two are not close. Elapsed time exceeds active review time in 43 of the 48 jurisdiction tables. Seattle reports 696 calendar days elapsed against 231 active. Shoreline 804 against 349. Bellevue 390 against 85. Across all tables the medians are 108 days elapsed and 65 active.
An authority that publishes one of these publishes the one it is measured against, and it is not the wait a builder experiences. Washington prints both, which is the only reason the gap is visible here at all — and nothing on this page establishes how wide it runs in the states that publish neither.
The gap is also not evenly spread. It is widest on construction permits and multifamily housing and narrowest on final subdivisions, which is a mechanism rather than a level — and the next section is where it opens.
What is counted
Elapsed, including hold
From the notice of complete application to the decision, with the days the file sat with the applicant still on the clock. This is the wait, and it is what this project measures.
What is usually published
Active review only
The same interval with the pauses removed. A defensible measure of agency throughput, and a house is not delivered faster because the clock stopped.
RegisterE-189confirmedE-190confirmedcheck these rows →
The measurementWhere the gap opens, and where it closes
Across 118 jurisdiction-by-permit-type cells, the median share of elapsed time spent with the clock stopped is 37%. It is not spread evenly.
Construction permitsNew building work. The class that puts housing on the ground.
54%
Multifamily housing permitsApartments. Design iteration is heaviest here and so is the hold.
44%
Preliminary subdivisionsDividing land. Substantial applicant work, less back-and-forth on drawings.
22%
Binding site plansA site-layout instrument, mostly administrative.
19%
Final subdivisionsThe recording step. Almost nothing sits with the applicant.
13%
Hold time exceeds half of elapsed time in 39 of the 118 cells. The ordering is the point: the gap is widest on construction permits and multifamily housing — the classes where an applicant is redrawing plans between review cycles — and narrowest on final subdivisions, which are close to a recording step. The understatement is worst for exactly the permits that put housing on the ground.
28 of the 118 cells record no hold time at all. Those are real zeroes in the return and they are carried rather than smoothed — though Commerce's own list of implementation problems notes that some permitting software cannot track when a review is paused, so an unknown number of them are systems that cannot see the gap rather than jurisdictions that do not have one.
RegisterE-194confirmedE-190confirmedcheck these rows →
The shapesFour ways a number stops short
Washington is one case. Of the four authorities whose definitions could be checked for this project, none publishes the elapsed wait as its headline figure — and each falls short of it differently, which is why a reader cannot learn one rule and apply it.
Form one
The clock stops
Applicant hold time is excluded. Defensible as agency throughput, and a house is not delivered faster because the clock was paused. Washington publishes both clocks, which is the only reason the size of this is known.
Form two
The endpoint arrives early
Portland, Oregon reports time to “approved to issue” — the point where, in the city's own words, review teams finish their work and await payment of final permit fees by the customer. It ends where the city's work ends, not where the permit does.
Form three
The deadline moves
England reports 91% of major applications decided within 13 weeks or within an agreed extended time. Against the statutory 13 weeks alone the figure is 19%, and 77% of major decisions involved such an agreement.
Form four
Nothing is computed
California collects every date and publishes no duration at all. The subtraction is left to the reader, and two in three of its jurisdictions file something no subtraction survives.
Every one of those figures is correct. That is what makes the problem durable: none of these authorities publishes a false number, each publishes an accurate answer to a question the reader did not ask. An authority that publishes one figure publishes the one it is judged on, and it is judged on the part of the process it controls. It becomes a problem one step downstream, when the figure is lifted into an argument about how long housing takes.
RegisterE-196supportedE-195confirmedE-157confirmedcheck these rows →
The recordThree things a reader of the source should know
None of these is a criticism of either state. Both publish more than any other state does. They are recorded because anyone repeating this work will hit them, and because two of the three would silently change the answer.
| The columns are labelled inconsistently | 45 of 48 | Forty-five of Washington's 48 jurisdiction tables print column 5 as “with pauses”. Arlington and Battle Ground print the two reversed. The data is consistent — column 5 is the larger figure in all 288 data rows — so the columns are read by position, not by label. Reading the labels swaps the elapsed wait for the agency clock in two jurisdictions. |
| The headline table is not a duration | Table 5 | The report's comparative ranking of all jurisdictions is median review days compared to the statutory deadline — a signed difference. Most of its bars are negative. A jurisdiction well inside a 170-day deadline and one well inside a 65-day deadline look identical. |
| The scope moved between years | HB 1935 | Building permits were removed from the definition of a project permit in 2025, so the reportable set shrank. The year-on-year improvement the report notes is measured across a changed denominator, and is not carried here as a change in permitting speed. |
RegisterE-191confirmedE-192confirmedcheck these rows →
The findingIt survived
This exercise was built to test one thing: whether compelling a jurisdiction to report its permitting timing produces a permit duration anybody can look up. In the only two states that compel it, the answer is no.
California holds the dates for 330,855 projects and publishes no statistic computed from them; the figures on this page exist because two published tables were joined, and two in three of its jurisdictions file something no duration can be computed from. Washington publishes a statistic and it measures a narrower thing than the phrase suggests — a class of permits that excludes building permits, on a clock that starts at the notice of complete application rather than at submittal.
That is a stronger version of what this project already recorded about construction duration, and it points the same way: the join exists, somebody stamped the application and somebody issued the permit. It is held and not published.
RegisterE-193supportedE-134confirmedcheck these rows →
The proposalNothing here needs a new survey
For housing that goes through the permit system, closing this gap requires no new data collection at all. A public body already holds the application date and the decision date for the same building, and publishes neither as a duration.
England is the clearest case: the department holds the dates for every planning application in the country and publishes no mean and no median, only compliance against a target that can be renegotiated after the clock has started. California is the same shape with the opposite symptom — it collects application, entitlement, permit and occupancy dates from every city and county in the state, publishes all of them, and computes nothing.
One
Publish the subtraction
California already has both dates in two tables it publishes separately. Joining them is arithmetic, and this page is the evidence that an outsider can do it. The state can do it better, because it can fix the filings rather than screen them out.
Two
Report elapsed, not variance
Washington already computes the elapsed wait and prints it in an appendix, then leads with the difference from a statutory deadline. Leading with the elapsed figure it already has would cost nothing and would make the series readable.
Three
State the definition beside the number
Which clock, which endpoint, whether hold time is in or out. Four authorities were checked here and no two measured the same interval. A duration collected against an unstated definition cannot be repaired later.
Four
Fix the filing, not the average
Two in three California jurisdictions file something no duration survives, and the defect is a property of the jurisdiction. A validation rule at submission would close more of this gap than any amount of analysis afterwards.
None of that is a research programme. It is publishing a subtraction, leading with a figure already computed, naming a definition, and validating a form. The join exists in every one of these jurisdictions — somebody stamped the application and somebody issued the permit. It is held and not published, and that is a choice rather than a limitation.
Housing that never enters the permit system at all — most of the Mexican owner-occupied stock, and the self-built share across much of the world — is invisible to every proposal above, because no filing requirement touches it. There a survey is the only instrument there is, and none of this substitutes for one.
RegisterE-193supportedE-196supportedE-157confirmedcheck these rows →
The workingEverything behind the figures above
Research data
Both extracts carry the columns the screen runs on — match rate, matched project count and identical-date share for every jurisdiction, whether or not a median is published for it. A reader who wants a different threshold can apply one. The measurement spec ships beside them because the definitions are most of this job: a number collected against the wrong definition is worse than no number, since it survives review looking like the thing it is not.
California already holds an occupancy date in the same table it holds the permit date, so application-to-occupancy is computable and is deliberately not computed here: the occupancy field is populated far more sparsely than the permit field, and the projects that have one are the ones that finished fastest, so the censoring that merely biases the permit measure would dominate an occupancy measure. Doing it properly needs a survival model rather than a subtraction. Either state could publish the duration directly and end the argument — California by computing what it already collects, Washington by reporting the elapsed figure it already prints as its headline instead of the difference from a deadline.
Three doorsWhere this sits, and what would settle it
01 · browse
Every figure on this page is a row with a source, a grade and a review date — including the rows that record what neither state publishes.
→ /evidence
02 · the other half
Thirty-eight countries, and not one that publishes approval, waiting and construction as a single series. The same absence, one scale up.
→ /international
03 · the concessions
The falsification statement, filed before any evidence was collected, with the clause that has already been tested and lost still on it.
→ /about