Research directory
Where California cities publish their permit data — and what state it is in
A portal is not a dataset. We ran the queries, recorded the last record in each store, and published the ones that publish nothing.
Written by HyreADU Research Desk Primary-source research and data analysis
Audited by HyreADU Research Desk Query reproducibility and retrieval-date audit
The finding
The useful question is not whether a California city has an open data portal. It is what date the last row in it carries. We queried each endpoint directly on 2026-09-05 rather than reading the portal’s own description of itself, and the spread is extreme.
San Francisco’s building permits dataset is the best in the state that we could reach. It holds 1,295,048 records, the most recent filed at 16:49 on 4 September 2026 — the day before retrieval — and, alone among everything we reached, it carries a dedicated adu field on every record. 3,269 records are flagged as ADUs.
Los Angeles publishes more permit datasets than anybody and keeps none of them current. Its largest, “LA BUILD PERMITS”, holds 375,150 records and stops on 19 May 2023 — 1,205 days before retrieval — while the Socrata catalog reports the dataset as updated on 29 January 2025.
A second dataset, published under the title “Building Permits Feb’15-Present”, holds 14,793 records and stops on 11 March 2015.
Los Angeles is nonetheless the only store we reached that names the contractor. Its schema carries Contractor’s Business Name, License #, License Type and License Expiration Date. San Francisco’s carries none of those.
Nobody we reached publishes both a contractor of record and a live feed. Publishing nothing is a finding, and we record it as one. San Mateo County runs a Socrata portal that returns zero results for “permit”.
San José and San Diego are not on the Socrata network at all. Every number on this page is a live API response, and every query is printed so you can re-run it.
This directory covers what we could reach by machine on one day; it is not a census of California’s 539 jurisdictions, and we do not present it as one.
Read this first
- The directory covers what we reached, not a census of California
California has 482 cities and 58 counties. We queried the endpoints we could construct a URL for.
Where a jurisdiction is absent from this page, the correct reading is “HyreADU did not reach it on 5 September 2026”, not “it publishes nothing”.
The three jurisdictions we do record as publishing nothing are recorded that way because a query against their own portal returned zero rows, which is a different and stronger claim.
- Web search was unavailable for this retrieval, which biases the sample toward Socrata
The federated Socrata discovery API lets us enumerate datasets on any Socrata-hosted domain without searching. It cannot see ArcGIS Hub, CKAN, Accela public portals, Tyler EnerGov or a PDF on a planning page — and a large share of California jurisdictions use one of those.
The sample here is therefore skewed toward the publishers who happen to use Socrata. That is a real limitation and we would rather name it than quietly present a partial list as a complete one.
- Catalog metadata is not evidence of freshness, and we stopped treating it as such
Every dataset in the Socrata catalog carries an
updatedAttimestamp. In the Los Angeles case that timestamp reads 29 January 2025 while the newest permit in the data is dated 19 May 2023.The timestamp records when the asset was last touched, not when the content last advanced. Any directory built on catalog metadata alone will report Los Angeles as reasonably current. It is not.
- A permit record is an application, not a building
Everything counted here is a permit record in a permitting system. A filed application may be withdrawn, a permit may be issued and never built, and one project can generate several records across building, electrical, plumbing and mechanical trades.
Nothing on this page is a count of completed accessory dwelling units. For the completions question see our permits against completions study, which is about exactly this gap.
Why a directory of the data, rather than the data
Anyone who has tried to answer a question about California housing production from municipal sources has met the same wall twice. The first time is finding the data at all.
The second, worse time is discovering that what you found stops three years ago, and discovering it only after you have built an analysis on it.
The second failure is the expensive one, because nothing about a stale dataset looks stale. It has a title with the word "Present" in it. It has a catalog entry with a recent timestamp.
It has an API that returns a 200 and a JSON array. The only way to know is to ask it for its maximum date, which takes one query and is the query almost nobody runs.
HyreADU analysis: so this page is deliberately not a data page. It is a directory with a freshness column, because the freshness column is the one that decides whether the rest of the row is worth anything.
We publish the query beside each figure so that the check costs you the same one request it cost us, and so that when a publisher fixes a feed you can prove it rather than take our word for the state of the world on one Saturday in September 2026.
The directory
One row per store we reached. “Last record” is the maximum date in the data itself, not the catalog metadata. Every figure in this table came back from the query printed in the last column, run on 2026-09-05.
| Publisher and dataset | Platform / resource id | Records | Last record in the data | Contractor of record | ADU identifiable | How we measured it |
|---|---|---|---|---|---|---|
| City and County of San Francisco — Building Permits | Socrata, data.sfgov.org, id i98e-djp9 | 1,295,048 | 4 September 2026, 16:49 (filed_date). One day before retrieval. | No. No contractor or license field in the schema. | Yes — a dedicated adu field on every record, values Y and N. | $select=count(1),min(filed_date),max(filed_date) on the resource endpoint. |
| City of Los Angeles — “LA BUILD PERMITS” | Socrata, data.lacity.org, id xnhu-aczu | 375,150 | 19 May 2023 (issue_date). 1,205 days before retrieval. Catalog claims updated 29 January 2025. | Yes — Contractor’s Business Name, Contractor Address, Contractor City, Contractor State, License #, License Type, License Expiration Date. | No flag. Would require parsing the free-text Work Description field. | $select=count(1),min(issue_date),max(issue_date) on the resource endpoint. |
| City of Los Angeles — “Building Permits Feb’15-Present” | Socrata, data.lacity.org, id 6q2s-9pnn | 14,793 | 11 March 2015 (issue_date). 4,196 days before retrieval, under a title containing the word “Present”. | Yes — same contractor and license fields as above. | No flag. | Same query shape. The title and the data disagree by more than eleven years. |
| City of Los Angeles — five further permit datasets | Socrata, data.lacity.org | Not counted | Not measured individually | Varies by dataset | No flag in any of them | Enumerated from the catalog: “Building Permits: New Housing Units” (cpkv-aajs), “Building Permits over 100K Valuation” (y5ik-mwat), “Apartment Building Permits over 100k” (n3m5-kxzy), and two datasets both titled “Eastside Building Permits 2018” (u8xs-sfb6, yrwh-n3zy). |
| County of San Mateo | Socrata, data.smcgov.org | Zero permit datasets | n/a | n/a | n/a | Federated catalog query for “permit” scoped to the domain returned resultSetSize: 0. The portal exists; the permit data is not on it. |
| City of San José | Not on the Socrata network | Not reachable this retrieval | Not measured | Unknown | Unknown | Catalog query scoped to data.sanjoseca.gov returned HTTP 404, meaning the domain is not federated with Socrata. The city publishes elsewhere; we could not construct the endpoint without web search. |
| City of San Diego | Not on the Socrata network | Not reachable this retrieval | Not measured | Unknown | Unknown | Catalog query scoped to data.sandiego.gov returned HTTP 404. A direct fetch of the documented dataset path also returned 404. |
Building permit publication by California jurisdiction, as measured by direct query on 2026-09-05. “Records” and “last record” are values returned by the publisher’s own API, not values reported in a catalog entry.
HyreADU calculation: the day counts in the “last record” column are ours — the number of days between the maximum date the endpoint returned and 5 September 2026. Everything else in the table is a value the API returned.
The San Francisco ADU flag, and why it is unusual
One field, on one dataset, in one city. It is the difference between counting ADU permits and guessing at them.
adu = Y flag, by filing year. The series runs from a single record in 2013 through a 2020 peak of 501, a trough of 120 in 2024, and 186 in 2025. The 2026 column is a partial year covering filings to 4 September 2026 and is drawn lighter for that reason. A further 78 flagged records carry no filing date and are excluded from every column. Chart: HyreADU Research Desk from a single grouped SoQL query against data.sfgov.org dataset i98e-djp9, retrieved 2026-09-05. The fourteen columns sum to 3,191, which with the 78 undated records is the 3,269 the ungrouped count returns. Reproducible from the year-by-year table below. What the field actually is
The dataset carries a field named adu alongside the usual permit apparatus — permit_number, permit_type, filed_date, issued_date, estimated_cost, existing_units, proposed_units, existing_use, proposed_use.
It takes the values Y and N. Grouping on it returns 3,269 Y and 1,291,779 N, which sum exactly to the dataset total of 1,295,048.
What it is not: a count of ADUs. It is a count of permit records flagged by San Francisco staff as relating to an ADU. One project can generate several. Some will have been withdrawn. It is nonetheless the only structured ADU signal in any California permit store we reached.
Why the alternative is so much worse
Without a flag, the only route to an ADU count is a text search of the work-description field for terms like "ADU", "accessory dwelling", "granny flat", "secondary unit", "junior ADU" and the dozen local variants — a method with no stable recall, no way to measure what it missed, and a false-positive rate that changes with whichever clerk typed the description.
HyreADU analysis: this is why the ADU research literature is so much thinner than the interest in the topic would suggest.
The unit of analysis is not reliably identifiable in the administrative record of most jurisdictions.
San Francisco solved it with one boolean column, which cost nothing and is worth more than any modeling exercise built on top of a text search.
What the shape of the series does and does not tell you
The series peaks in 2020 at 501 filings and falls to 120 in 2024 before recovering to 186 in 2025. It is tempting to read that as a demand collapse.
We are not going to read it that way from this data alone, because a filing count is sensitive to at least three things that have nothing to do with demand: a change in how staff apply the flag, a change in how many separate records a single project generates, and a backlog in getting records into the published extract.
What we will say is narrower and safe: the number of San Francisco permit records flagged as ADUs fell by 76 per cent between 2020 and 2024 and recovered by 55 per cent in 2025.
Whether that is fewer ADUs or a different clerical practice is a question this dataset cannot answer, and answering it would require the department, not the portal.
The San Francisco series, as numbers
The chart above, in a form you can check against your own query. Every row is a group returned by the same request.
| Filing year | Permit records flagged adu = Y | Change on prior year | Note |
|---|---|---|---|
| 2013 | 1 | — | First flagged record in the store. |
| 2014 | 1 | 0 | The flag is present but effectively unused. |
| 2015 | 39 | +38 | San Francisco’s local ADU legalization programs begin in this period. |
| 2016 | 267 | +228 | First full year of substantial use. |
| 2017 | 381 | +114 | Highest pre-2020 year. |
| 2018 | 355 | −26 | |
| 2019 | 367 | +12 | The year the 2019 state package (AB 68, AB 881, SB 13) was chaptered; it took effect 1 January 2020. |
| 2020 | 501 | +134 | Series peak. First year the 2019 state package was operative. |
| 2021 | 341 | −160 | |
| 2022 | 306 | −35 | |
| 2023 | 231 | −75 | |
| 2024 | 120 | −111 | Series trough — 76 per cent below the 2020 peak. |
| 2025 | 186 | +66 | Recovery of 55 per cent on 2024. |
| 2026 (partial) | 95 | n/a | Records filed to 4 September 2026. Not comparable to a full year. |
| Total | 3,269 | Includes 78 flagged records with no filing date, which appear in no year above. |
San Francisco permit records flagged as accessory dwelling units, by filing year. Source: data.sfgov.org dataset i98e-djp9, grouped SoQL query, retrieved 2026-09-05. The year-on-year change column is a HyreADU calculation.
Not a count of accessory dwelling units built, permitted, or completed. A permit record is an application in a permitting system. See the limitations block at the top of this page.
How to run this check yourself
Four requests. No credentials, no tooling, no account. If you take one thing from this page, take this method rather than these numbers — the numbers will age and the method will not.
- 1 Ask the catalog what exists on the domain
The Socrata federated discovery API answers
https://api.us.socrata.com/api/catalog/v1?q=building+permits&domains=data.lacity.orgwith every dataset on that domain matching the term, including each dataset’s four-by-four resource id and its full column list. A 404 means the domain is not federated with Socrata, which is itself a useful answer: the publisher is on ArcGIS Hub, CKAN, or a vendor portal. - 2 Ask the dataset for its own boundaries before you download anything
Append
$select=count(1) as n, min(filed_date) as first, max(filed_date) as lastto the resource JSON endpoint. One request, three numbers, and it settles whether the rest of the exercise is worth doing. This is the step that separates a live feed from an abandoned one, and it is the step routinely skipped. - 3 Compare the maximum date against the catalog timestamp, and believe the data
Where the two disagree, the catalog timestamp is describing asset maintenance and the maximum date is describing content. Los Angeles is the worked example: catalog says January 2025, data says May 2023. Report the data date.
- 4 Read the column list before you write the query
The catalog response carries
columns_name. That is where you discover that San Francisco has anadufield and no contractor field, and that Los Angeles has seven contractor and license fields and no ADU field. Neither fact is discoverable from the dataset’s title or description, and both change what questions the store can answer. - 5 Record the ones that publish nothing, in the same table
A directory that silently omits the non-publishers overstates coverage. San Mateo County returning zero permit datasets from its own portal is a result, and belongs on the page beside the cities that publish well.
Two stores, opposite strengths, no overlap
San Francisco publishes the unit, not the builder
1,295,048 records, current to the day before retrieval, with an adu flag, estimated_cost, revised_cost, existing_units, proposed_units, existing_use, proposed_use, filed_date, issued_date, approved_date, supervisor_district and a geocoded location.
What it cannot answer: who built it. There is no contractor name, no CSLB license number, no applicant business field in the schema. A question of the form "which contractors are doing the ADU work in San Francisco" cannot be asked of this dataset at all.
Los Angeles publishes the builder, three years late
375,150 records with Contractor’s Business Name, License #, License Type, License Expiration Date, Contractor Address, Valuation, # of Residential Dwelling Units, Floor Area under both the zoning and building code definitions, and Census Tract.
What it cannot answer: anything about the present. The newest record is from 19 May 2023. And it carries no ADU flag, so isolating accessory dwelling units means text-matching the work description with no measurable recall.
HyreADU analysis: the two schemas are close to complementary, and neither jurisdiction publishes what the other does.
There is no California store we reached from which you can answer both "how many ADUs were permitted last quarter" and "who holds the license on them".
Our own permit-to-contractor index exists because of exactly this gap, and inherits its limits.
Terms this page uses precisely
- Four-by-four (resource id)
- Socrata’s permanent identifier for a dataset — four characters, a hyphen, four characters, as in
i98e-djp9. It survives renaming and retitling, which is why every row in the directory above cites the id rather than only the title. Two Los Angeles datasets share the title “Eastside Building Permits 2018” and are distinguishable only by id. - SoQL
- The SQL-like query language Socrata endpoints accept as URL parameters —
$select,$where,$group,$order,$limit. Every figure on this page came from a SoQL expression appended to a public JSON endpoint, with no key and no client library. - Catalog freshness against data freshness
- Two different things that a directory routinely conflates. Catalog freshness is the
updatedAttimestamp on the asset record and moves whenever anything about the asset is touched. Data freshness is the maximum date in the rows. Where they disagree, only the second one tells you whether the dataset describes the present. - Federated domain
- A domain whose datasets are indexed by Socrata’s cross-publisher discovery API, and therefore enumerable without a search engine.
data.sfgov.org,data.lacity.organddata.smcgov.orgare federated.data.sanjoseca.govanddata.sandiego.govreturned 404 from the catalog, meaning they are not — not that they publish nothing.
What we could not verify
Every non-Socrata portal in California. Web search was unavailable for the whole of this retrieval, so we could enumerate only what the Socrata federated catalog could see.
ArcGIS Hub sites, CKAN instances, Accela and Tyler EnerGov public portals, and jurisdictions that publish a monthly PDF are all invisible to that method.
Sacramento, Marin County, San José and San Diego are therefore unmeasured on this page rather than absent from the world, and we will not characterize their publication practice on the strength of a failed URL guess.
San Diego’s dataset path. We attempted a direct fetch of data.sandiego.gov/datasets/building-permits/ and received HTTP 404. That is consistent with the city having reorganized the path, not with the city having withdrawn the data. We record the attempt and the response and nothing further.
The five remaining Los Angeles permit datasets. We enumerated them from the catalog but measured only two of the seven for record count and maximum date. The five unmeasured ids are printed in the directory so that anyone can complete the exercise in five requests.
Whether the San Francisco ADU flag is applied consistently over time. The 2024 trough could be a change in filing volume or a change in clerical practice, and the published data contains nothing that distinguishes them.
We have said so on the page rather than picking the more interesting explanation. Resolving it needs the department, not the portal.
Field-level completeness anywhere. A field existing in a schema is not the same as a field being populated.
We confirmed that Los Angeles carries a contractor license column; we did not measure what share of its 375,150 rows have a value in it.
That measurement belongs in the next edition, and is one query per column.
Questions
Which California city publishes the best building permit data?
Is Los Angeles building permit data current?
How many ADU permits has San Francisco filed?
Can I find out which contractor built a particular ADU?
Is there a statewide California building permit dataset?
Why does a dataset’s catalog entry say it was updated recently when the data is years old?
updatedAt field advances when the asset is touched — metadata edited, schema adjusted, a pipeline run that wrote no new rows. The maximum date in the rows is the only thing that tells you how far the content reaches. The Los Angeles case on this page is the clean demonstration: metadata January 2025, data May 2023.Does San Mateo County publish building permit data?
How often should this directory be re-checked?
Written and audited by
HyreADU Research Desk
Primary-source research, data analysis and fact checking
We are a research desk, not a builder. We read the permit extract, the statute, the HCD return or the fee schedule ourselves, and publish each figure with its source and retrieval date.
Where a number cannot be traced to a primary source, we leave it out and say what we could not verify. Our store-based claims cover California only.
- CA
- the only state this desk will make store-based claims about
- 5
- jurisdictions with extracted ADU permit evidence
- 735
- CSLB-verified companies in the California store
- 0
- national claims from a one-state store
How this desk works
- Primary sources only. Permit counts come from the city or county that issued the permit. Production counts come from HCD’s Annual Progress Report. Rents come from HUD or the Census. We do not cite an article that cites a source; we download the source and compute the figure ourselves.
- This is a California site. The company store is 734 California firms and one New Mexico firm. Permit evidence exists for five named jurisdictions: Los Angeles, San Francisco, Sacramento, San José and unincorporated Marin. A number from that store is titled to those places, never to the United States.
- A permit is not a completion, and a license is not an ADU grade. California licenses no ADU classification. Being named on an ADU permit is evidence of engagement in that jurisdiction, not of quality, completion, or work anywhere else. Owner-builder permits are excluded from contractor counts.
- Calculation is labeled as calculation. Figures we derive are never presented as something HCD, HUD, the Census or a city published. Terner Center research is cited as Terner’s, never restated as ours.
- We do not design, permit or build ADUs, and we take no payment for placement, ranking or a favorable mention. Pages that look like rankings are not: they publish public-record counts and let the reader decide.
- Nothing here is legal, tax or financial advice. Zoning, underwriting and appraisal practice vary by jurisdiction, lender and appraiser. The useful next step on a specific lot is the planning counter and a licensed professional.
Data as of All counts, minimum and maximum dates retrieved by direct Socrata SoQL query against each publisher’s own endpoint on 2026-09-05; catalog metadata retrieved from the Socrata federated discovery API on the same date. Authorship on this site is organizational: the analysis belongs to the desk rather than to a named individual, and we do not publish credentials we do not hold.
Our editorial policy sets out how we source, date and correct what we publish.
Sources & retrieval dates
- City and County of San Francisco — Building Permits (Socrata resource i98e-djp9) , Direct SoQL queries on 2026-09-05. Record count 1,295,048; min filed_date 1901-03-10T00:07:00.000; max filed_date 2026-09-04T16:49:46.000. Grouped on the adu field: Y = 3,269, N = 1,291,779. Grouped by year on filed_date where adu = Y: 2013:1, 2014:1, 2015:39, 2016:267, 2017:381, 2018:355, 2019:367, 2020:501, 2021:341, 2022:306, 2023:231, 2024:120, 2025:186, 2026:95, undated:78. Field list confirmed from a single-record fetch and includes permit_number, permit_type, filed_date, issued_date, estimated_cost, revised_cost, existing_units, proposed_units, existing_use, proposed_use, adu, supervisor_district and location; it contains no contractor or license field. Retrieved 2026-09-05.
- City and County of San Francisco — ADU-flagged permits grouped by filing year , The exact query behind the chart and the year table. Returns fifteen groups summing to 3,269, of which one group has a null year and holds 78 records. Retrieved 2026-09-05.
- City of Los Angeles — LA BUILD PERMITS (Socrata resource xnhu-aczu) , Direct SoQL query on 2026-09-05. Record count 375,150; min issue_date 2015-11-03T00:00:00.000; max issue_date 2023-05-19T00:00:00.000. The federated catalog entry for the same resource reports updatedAt 2025-01-29T03:00:02.000Z and createdAt 2015-10-13T14:57:16.000Z. Retrieved 2026-09-05.
- City of Los Angeles — Building Permits Feb’15-Present (Socrata resource 6q2s-9pnn) , Direct SoQL query on 2026-09-05. Record count 14,793; min issue_date 2015-02-01T00:00:00.000; max issue_date 2015-03-11T00:00:00.000. The catalog entry reports updatedAt 2025-01-29T02:44:52.000Z. Retrieved 2026-09-05.
- Socrata federated discovery API — data.lacity.org permit dataset enumeration , Returned seven permit datasets with full column lists: LA BUILD PERMITS (xnhu-aczu), Eastside Building Permits 2018 (u8xs-sfb6), Building Permits: New Housing Units (cpkv-aajs), Building Permits Feb’15-Present (6q2s-9pnn), Eastside Building Permits 2018 (yrwh-n3zy), Building Permits over 100K Valuation (y5ik-mwat), Apartment Building Permits over 100k (n3m5-kxzy). Contractor and license fields confirmed present in the column lists. Retrieved 2026-09-05.
- Socrata federated discovery API — data.smcgov.org permit query , Returned resultSetSize 0 with an empty results array. San Mateo County’s open data portal is federated with Socrata and carries no dataset matching “permit”. Retrieved 2026-09-05.
- Socrata federated discovery API — data.sanjoseca.gov and data.sandiego.gov , Both domain-scoped catalog queries returned HTTP 404, indicating the domains are not federated with the Socrata discovery network. This establishes that we could not enumerate them, and establishes nothing about what those cities publish by other means. Retrieved 2026-09-05.
- City and County of San Francisco — Building Permits single-record schema fetch , One-row fetch used to enumerate the field list verbatim, confirming the presence of the adu field and the absence of any contractor, license or applicant-business field. Retrieved 2026-09-05.
Check the freshness before you build on the feed
One request tells you whether a permit dataset describes the present or 2015. The queries on this page are public, keyless and take a second each — run them against your own jurisdiction before you trust a number that came out of it.
HyreADU does not design, permit or build accessory dwelling units. This page is a record of what specific public API endpoints returned to specific public queries on 5 September 2026, with each query printed so it can be re-run.
Figures will change as publishers update their stores, and a jurisdiction absent from this directory was not reached by our method rather than proven to publish nothing.
Counts of permit records are counts of administrative records, not of buildings: an application may be withdrawn, a permit may be issued and never built, and one project can generate several records. Nothing here is legal, tax or financial advice.