California permits thousands of ADUs. How many get built?

What 57.3 percent is, and what it refuses to be

The political story of California ADUs is a permit story. State law stripped parking minimums, setback traps and owner-occupancy rules; cities started issuing building permits in volume; the permit count became the measure of the reform.

Table A2 is the file that can test the next sentence — the one that treats those permits as housing. It can, if the two columns are not collapsed.

Over 2018–2025 the event-year series records 178,495 permitted ADU units and 102,356 completed ADU units. The cumulative ratio is 57.3%. Exclude the four unusable reporters and it is 57.7% on 177,426 permits and 102,352 completions.

The exclusion moves the statewide number by four tenths of a point. It is in the method because counting Huntington Beach’s 495 permits as 0 completions would be a claim about production that the city’s own APR cannot support.

The same-year ratio — this year’s certificates over this year’s permits — rose from 26.4% in 2018 (12,050 permits, 3,186 completions) to 65.5% in 2024 and 70.4% in 2025 (32,126 permits, 22,630 completions).

That climb is what a maturing pipeline looks like: early years are almost all new permits and almost no certificates; later years inherit certificates from the bulge. 2020’s 61.5% same-year ratio sits on a nearly flat permit year (12,992 then 13,116) while completions rose from 5,957 to 8,065 — catch-up in the certificate column, not a jump in new permits. 2021’s dip to 47.9% is the same arithmetic in reverse: permits jumped from 13,116 to 21,163 while completions rose only from 8,065 to 10,147.

HyreADU analysis: the honest sentence is that California is permitting ADUs faster than it is recording them as ready, that the gap has narrowed as a share of the annual flow, and that the remaining gap is a mixture of construction lag, abandonment, certificate non-reporting and identifier change — in unknown proportions.

A page that picked one of those four and called it “the” completion rate would be the thing this study exists not to publish.

The same-year ratio is a lag clock

If every permitted ADU finished in the year it was permitted, the same-year ratio would sit near 100 percent and the cumulative ratio would too. Neither does. The same-year series starts at 26.4% and ends at 70.4%. The bars below are that series, with the two counts that produce it.

The median lag among matched projects is 1 year (25th percentile 0, 75th percentile 1).

That is why a 2025 same-year ratio of 70.4% can coexist with a cumulative ratio of 57.3%: most certificates in 2025 belong to 2023–2024 permits, and most 2025 permits have not yet reached a certificate.

Treating 70.4% as “70 percent of 2025 ADUs got built” would reverse the clock.

The APR-year total double-counts the same permit

Jurisdictions re-file open projects in later Annual Progress Reports. Summing NO_BUILDING_PERMITS by the year the APR was filed therefore counts the same physical permit every year it is still open.

Over 2018–2025 that naive sum is 213,732 permit units against 178,495 unique event-year units — 35,237 double-counted units, 16.5% of the naive total.

Completions double-count too, more mildly: 108,252 naive against 102,356 event-year, which is why the naive cumulative ratio (50.6%) looks worse than the event-year ratio.

It is worse because the denominator is inflated more than the numerator.

2018 is the year that shows the mechanism. The 2018 APR-year permit sum is 9,000; the event-year sum, using the date on the permit, is 12,050 — 3,050 higher.

Later APRs back-filled 2018-dated permits that the 2018 filing had not yet recorded.

From 2019 onward the APR-year sum runs ahead of the event-year sum, by 9,721 units in 2025 (41,847 naive against 32,126 event-year).

A dashboard that charts “ADU permits by APR year” is, from 2019 on, a chart of re-reporting as well as of new permits.

HyreADU analysis: the double-count is large enough to change a headline and small enough to hide in a downloaded CSV that looks official. 16.5% of the naive permit total is not a rounding error.

It is why this page never sums by APR_YEAR, and why a contemporaneous snapshot (Terner’s 2020 paper, below) will not match an event-year reconstruction of the same statute years from a later extract.

ADUs are a quarter of HCD-reported permits. Census BPS cannot see that.

Inside HCD’s own permit file, ADUs went from 9.1% of reported units in 2018 to 26.5% in 2024 and 26.1% in 2025.

That is the reform, measured as a share of the state’s reported housing permits rather than as a raw ADU count.

Single-family detached permits in the same HCD series fell from 45,928 in 2018 to 35,424 in 2025 while ADU permits in that comparison series rose from 11,978 to 31,460.

The Census Bureau’s Building Permits Survey has no ADU category. BPS has no ADU category. Detached ADUs may be counted as 1-unit new residential construction; garage conversions and many internal ADUs may not be counted as new units at all.

This is a cross-check of statewide housing-permit levels, not a validation of the ADU count. We still put the two series in one table because a statewide housing-permit total that cannot see ADUs will drift from a statewide housing-permit total that can.

HCD’s all-type permit count runs above BPS every year in this window — 17,777 units higher in 2018 (1.16×) through 16,599 in 2025 (1.16×), peaking at 25,532 in 2023 (1.23×).

Adding HCD single-family detached to HCD ADU and comparing that sum to BPS 1-unit is a rough cross-check, not a reconciliation: −925 in 2018 and +9,145 in 2025.

The sign change is consistent with ADUs entering a 1-unit Census count only when they look like new houses, but this file cannot prove that.

HyreADU analysis: the mismatch is a classification problem, not a reason to prefer whichever number is larger.

The ADU share of HCD (9.1% → 26.5% / 26.1%) is the figure this page will defend, because both the numerator and the denominator come from the same APR.

Once a permit can be joined to a certificate, the lag is short

A project is a unique (jurisdiction, identifier) pair. The permit year is the earliest building-permit date; the completion year is the earliest certificate-of-occupancy (or other form of readiness) date.

Unusable reporters are excluded from the cohort so their blank completion fields are not treated as unfinished projects. Identifier matching is not a unique project ID issued by HCD; it is the best join the file supports.

Among 81,853 matched projects the lag distribution is tight. 24,077 (29.4%) receive a certificate in the same calendar year as the permit.

Another 40,486 complete the next year, which is why 78.9% are done within one year and 93.9% within two (12,264 more at t+2). 4,974 take three or more years.

The median is 1 year; the 25th percentile is 0; the 75th percentile is 1. 52 projects show a certificate dated before the permit (a legalization, a data error, or both).

They are 0.1% of matches — visible, and too few to move a median.

That is a different picture from the 57.3% window ratio, and it is supposed to be. The lag table is conditioned on a successful join.

Projects that never match — still under construction, abandoned, completed without a reported certificate, or filed under a changed identifier — are not in it. Conditional on appearing in both columns, California ADUs are not decade-long jobs. They are mostly one-year-to-the-next jobs.

The production gap lives in the unmatched residual, not in a long tail of matched construction.

The 2018 cohort is not a census of what got built

Follow a permit year forward and the observed match rate is the share of that year’s permitted units whose identifier later appears on a certificate in Table A2.

For the 2018 cohort that rate is 47.1% after 7 years of follow-up: 5,687 matched units out of 12,074, leaving 6,387 unmatched. Same-year observed completion for that cohort is only 9.9%; by t+2 it is 40.0%.

Rates are observed matches in Table A2, not a physical completion census. Unmatched units include projects still under construction, abandoned projects, and completed units whose certificate was not reported or whose identifier changed.

Later cohorts, with less follow-up, already show higher two-year observed matches than 2018 ever did. The 2022 cohort is at 62.4% after 3 years, and 57.3% by t+2. The 2023 cohort is at 57.8% by t+2.

If unmatched were simply “not yet finished,” the oldest cohort would be the most complete. It is not. 2018 is the lowest observed rate in the table despite the longest follow-up.

That is the opposite of a pure lag story, and it is why this page will not call 47.1% an abandonment rate. Identifier quality, reporting practice and legalization coding are in the residual alongside units that never rose.

Restrict the unmatched question to cohorts that have had at least two years to complete and the residual is still large: 46,944 unmatched units out of 114,651 permitted in 2018–2023 (40.9%).

The 2024 and 2025 cohorts are too young to belong in that sentence — 2025 has 0 years of follow-up and an observed match of 16.1%, which is almost entirely the same-year slice.

Reading that row as “16.1% of 2025 ADUs got built” would be a category error.

HyreADU analysis: the matched-lag table and the cohort-match table are doing different jobs. The first says: when both ends of the pipe are visible, the typical ADU is a one-year project.

The second says: a large minority of permitted units never become visible at the far end of this file, even after years, and we cannot see which of the four reasons applies.

Anyone using Table A2 as a construction census is asking a project-level question of a reporting file that was built for housing-element accounting.

Four jurisdictions whose completion fields cannot be read as production

A city that reports hundreds of ADU permits and no certificates can mean two things: nothing got built, or the certificate column was not used. Treating those as the same thing would produce a “zero-completion” ranking that is actually a ranking of APR hygiene.

Huntington Beach (Orange County, U1) reported 495 ADU permits across 7 years and 0 ADU completions; Completion fields for ADUs are unused across multiple reporting years.

The jurisdiction also reported zero certificates of occupancy for every other housing type in Table A2. 42.3% of its ADU rows have an empty tracking ID. Saratoga (Santa Clara County, U2) reported 467 permits across 8 years and 4 completions in a single year (0.9%).

Completions are so sparse relative to a large, multi-year permit series that treating the ratio as production would present non-reporting as non-building.

West Sacramento (Yolo, U1) and Arvin (Kern, U1) reported 65 and 42 ADU permits respectively, each with 0 ADU completions, while filing 116 and 228 certificates of other housing types — so the gap is ADU-specific, not a blank CO column.

Together the four account for 1,069 of 178,495 statewide permit units and 4 of 102,356 completions. The statewide ratio moves from 57.3% to 57.7% when they are set aside.

The point of naming them is not that arithmetic. It is so that no sentence on this page can be quoted as “Huntington Beach completed zero ADUs.”

Twenty large jurisdictions, sorted by permit volume — not by ratio

The table below is the twenty usable jurisdictions with the most ADU permits in 2018–2025. It is sorted by permits because volume is what the file measures cleanly. The ratio column is published so a reader can see the range.

It is not a score, and it is not quality. A high ratio can mean fast construction, aggressive certificate reporting, a draining pipeline, or some of each.

A low ratio can mean slow construction, a certificate column that started late, a 2025 filing that is still thin, or identifier practice that breaks the join.

This desk will not sort California cities by that mixture and call it a league table.

Los Angeles alone accounts for 49,556 of 178,495 statewide permit units (27.8%) and 29,529 of 102,356 completions (28.8%), with certificates in all eight years. The statewide series is, to a first approximation, Los Angeles plus a long tail. The twenty rows together are 95,839 permits (54.0% of the usable statewide total).

San Diego’s cumulative ratio is 29.1% on 9,153 permits and 2,667 completions, with certificates in five of eight permit years.

Unincorporated Santa Barbara County sits in a similar band (26.8%) with certificates in all eight years.

Those are not the same reporting pattern, and they are not a reason to tell a homeowner in either place that the city “doesn’t finish ADUs.” Long Beach’s 78.6% on eight years of both columns is the other end of the same warning: a high ratio is not a recommendation.

Ten places, year by year: where the calendar-year ratio breaks

The focus set is Los Angeles, San Diego, San José, San Francisco, Sacramento, Oakland, Long Beach, Fresno, unincorporated Los Angeles County and unincorporated San Diego County — the large markets a homeowner actually asks about, plus Fresno as a Central Valley check. Each cell is permit units / completion units in that calendar year.

Los Angeles issued 7,039 ADU permits in 2018 and recorded 1,424 completions that year; by 2025 those figures are 7,439 and 5,888. The city’s cumulative ratio (59.6%) sits next to the statewide 57.3% because Los Angeles is a large fraction of the statewide file, not because every city looks like Los Angeles.

San Diego recorded 0 ADU completions in 2018, 0 in 2019 and 0 in 2020, against 216, 611 and 491 permits. Completions then appear: 327 in 2021 through 851 in 2024, before 2025 falls to 229 completions against 2,165 permits.

Five completion years out of eight permit years is in the top-20 table. The year-by-year row is what that looks like. It is a reporting-and-lag pattern. It is not a finding that San Diego failed to build 6,486 ADUs.

Certificates can exceed same-year permits, which is what a lag plus a draining pipeline produces. Long Beach in 2025: 644 permits, 649 completions.

San Francisco in 2024: 230 / 249; in 2025: 171 / 218. Oakland in 2020, 2023 and 2024 likewise. A same-year ratio over 100 percent is not over-production. It is the earlier bulge arriving.

Fresno is in the focus set with 280 permits and 78 completions (27.9%). That is a small series. A ratio on 280 units does not carry the same weight as a ratio on 49,556, and we will not write the sentence that pretends it does.

Oakland’s 2018 cell (252 permits, 0 completions) is the same class of warning at the start of a series: a zero in year one is how a certificate column begins, not how a city is graded.

What Terner already measured, and what this page adds

The Terner Center for Housing Innovation at UC Berkeley owns much of the homeowner-facing ADU evidence in California — cost, finance, occupancy, and the early production counts. This page does not restate that work as ours.

It complements it on the cut Terner flagged and did not, in the 2020 paper, follow all the way through a later extract: permits versus certificates, on a de-duplicated event-year basis, with unusable reporters named.

Reaching California’s ADU Potential (August 2020) reported 5,911 ADU permits and 1,984 completions in 2018, and 15,571 permits and 6,668 completions in 2019. Terner reported contemporaneous APR snapshots.

Later APRs backfill earlier permit dates, so our 2018 event-year permit count is higher than Terner’s 5,911. The directional finding — completions lag permits — is the same.

Our 2018 event-year count is 12,050 permits and 3,186 completions; our 2019 event-year count is 12,992 and 5,957. Terner’s 2019 permit figure sits next to our naive APR-year 2019 total of 15,064, not next to the de-duplicated event-year total. Different vintage, different grouping, same directional finding: certificates lag permits.

Terner’s 2024 coastal-zone analysis measures a different clock. Average days from planning application to building permit, 2018–2022, inside vs outside the Coastal Zone in Los Angeles, Orange and San Diego Counties.

In Los Angeles County they report 260 days inside the zone and 147 outside; in Orange County, 233 and 101; in San Diego County, no statistically significant difference. Terner measures the application-to-permit clock. This study measures the permit-to-occupancy clock.

They are consecutive segments of the same pipeline, not competing estimates of the same number. The permit-times study is where application-to-permit belongs on this site.

This page does not compete with Terner’s days, and it does not convert our one-year median lag into a day count the file will not support.

HyreADU analysis: the useful division of labor is that Terner measured the homeowner barrier (cost, finance, time-to-permit, who lives in the unit) and we measured the production-accounting barrier (what share of the permit surge shows up as a certificate, how long the joinable projects take, and which cities cannot be read). Collapsing those into one “ADU success rate” would undo both.