How HyreADU research works — the method every page here is held to

What this desk is, stated plainly

A methodology page is the easiest place on a website to overclaim, so this one starts with the things that are not true.

The HyreADU Research Desk is organizational, not personal. There is no named researcher behind these pages, no photograph, no biography, no list of degrees and no count of years in the field.

Original research on this site is attributed to the desk as an organization and to nothing else. Where a page needs expert commentary, it is sourced from a named, verifiable person outside this site and cited to them.

It holds no credentials and no affiliations. It is not a licensed design or engineering practice, it does not practice law, it is not accredited by any body, and it is not a member of any research association.

It has no laboratory, no field survey operation and no proprietary panel of respondents. Every number on this site was either computed from a file somebody else published, or published by somebody else and cited.

HyreADU does not design, permit or build accessory dwelling units. The site expects to earn revenue by connecting homeowners with ADU professionals, and that relationship is disclosed wherever it applies.

Nobody can buy a placement, a ranking or a favorable description here. Pages that look like rankings are not: they publish public-record counts and let the reader decide.

What the desk does have is narrow and checkable. It downloads primary files, computes figures from them with a stated method, and publishes both the figure and the file. That is the whole claim, and the rest of this page is how it is held to it.

The sourcing rule: a document, and the date we opened it

Two requirements, applied to every factual claim on the site. Neither is negotiable, and together they are the reason several pages here are shorter than their competitors.

First: the source is the document, not an article about the document. If a page states a statutory requirement, the citation is the section on the Legislative Counsel’s own service, not a law-firm summary of it.

If a page states a permit count, the citation is the municipal open-data resource, not a news story quoting it. This is not fastidiousness.

The single most common failure in ADU material is a chain of citation that terminates in a builder’s brochure, and cutting the chain to length one is the only reliable defense against it.

Second: every source carries the date it was retrieved. Not the date of the source — the date this desk opened it.

Those are different facts and the second one is the one a reader needs, because a municipal fee schedule from 2024 that we read last week is a different quality of evidence from the same schedule read two years ago.

Retrieval dates appear next to the source in the source list, in the "data as of" line at the top of every study, and inside dataset files as a header comment.

Where those two requirements cannot be met, the claim does not ship. That has real consequences on this site and they are visible: the ordinance tracker publishes a coding schema and two jurisdictions rather than a statewide table, because codified municipal text could not be retrieved; the fee work publishes no statewide fee table for the same reason.

The download shelf carries no fee file, and the page says why in the section on what is deliberately missing.

Primary is preferred in a defined order. Current statute and chaptered bill text; the issuing agency’s own dataset or return; the jurisdiction’s own published page; a federal statistical series; peer-reviewed or institutional research cited to the institution; and, last and always labeled, a trade survey or index.

A source further down the list is not barred — it is labeled, and it is never used to contradict one further up.

Fact, calculation, analysis, recommendation

Fact, calculation, analysis and recommendation are four different kinds of claim and they carry four different burdens of proof. Blurring them is how a research site becomes a marketing site without anyone deciding to.

A source fact is quoted or closely paraphrased from a document we retrieved, and it reads that way: “Government Code § 66317(a)(3) provides that if the local agency has not approved or denied the completed application within 60 days, the application shall be deemed approved.” The reader can check it in one click.

A HyreADU calculation is a figure we computed from somebody else’s file, and it says so: “HyreADU analyzed the LADBS building-permit extract and calculated a median of…”.

It is never written as though the agency published it. The distinction matters because an agency figure has an agency’s methodology and review behind it, and ours has ours.

A HyreADU analysis is an argument: a reading of what a pattern most likely means, an explanation of why two sources disagree, an inference about the shape of a distribution.

It is always marked — “HyreADU analysis”, “the most defensible reading is”, “this suggests” — and it is always separable from the data underneath, so a reader who accepts our numbers can reject our interpretation. That is the point of publishing the file.

A recommendation is a suggested next step, and on this site it is almost always procedural: ask the agency in writing, check the section on leginfo yourself, take the state floor table to the counter. It is never a conclusion about whether a specific project is lawful, affordable, financeable or a good idea.

The rule that ties them together: we never present the desk’s interpretation as something a source stated, and we never say HyreADU collected data when HyreADU analyzed somebody else’s. Those are the two failures that would make everything else on this page worthless.

The California limit, and why it is written into the arithmetic

The California scope limit is the constraint that most often decides what a page is allowed to say, so it is stated as a rule rather than a caveat.

HyreADU is a California research program. The company store holds 735 CSLB-verified companies, of which 734 are Californian and one is not. Extracted ADU permit evidence exists for five named jurisdictions: the City of Los Angeles, San Francisco, Sacramento, San José and unincorporated Marin. That is the entire empirical base this desk owns.

A figure computed from that base is titled to those places and no others. A median from the Los Angeles permit extract is a Los Angeles median. A distribution across five jurisdictions is a five-jurisdiction distribution. Neither is a California figure, and neither is remotely a national one.

Where a page needs to say something about the United States, it rests on federal or state agency data — the Census, HUD, HCD, BLS — and it is cited to the agency, never to us.

The counted record is published rather than described. The research-desk block at the foot of every study carries four numbers: California as the only state we will make store-based claims about, five jurisdictions with extracted permit evidence, 735 CSLB-verified companies, and zero national claims from a one-state store.

Those numbers are counted from what the site actually holds. They stay small until the work is done, because the block is the only credential this desk has and inflating it would destroy the one thing it is for.

The same limit applies to time. A feed that stops is a feed that stops: the Los Angeles legacy contractor extract ends on 19 May 2023 by construction of the feed, not by choice, and pages that rest on it say so rather than implying currency they do not have.

Retrieval walls, and why we publish them

A significant amount of the material this work needs is public in principle and unreachable in practice. That fact is published rather than worked around.

The recurring obstacles are specific. The platform hosting most California municipal codes returns HTTP 403 to automated requests, so codified municipal ADU standards are frequently unobtainable. Some agency sites are intermittently unavailable, occasionally with a maintenance notice, occasionally without. Some license-detail services reject automated requests outright. Some municipal open-data resources have simply stopped refreshing.

When a wall is hit, three things happen and none of them is a substitution. The page states what was sought, what the obstacle was, and the date.

The affected cells stay empty rather than being filled from a commercial ADU site or an aggregator. And the page ships with whatever was actually verified, at whatever length that turns out to be.

This is the single most consequential rule on the site, because the alternative is always available and always looks better.

There is a version of the ordinance tracker with a row for every California jurisdiction, assembled from secondary sources, and it would be the most linked page here.

It is not published, and the tracker says in its own words that it has read two jurisdictions and says two.

A wall is not permanent and is not treated as such. Where a source becomes reachable, the page is re-run and the retrieval date changes. The record of what could not be reached stays, because a reader comparing two editions is entitled to know which cells were filled later and why.

Corrections, updates and decay

Research about a moving regulatory target decays. The policy is designed around that rather than against it.

Corrections are made in place. Where something here is wrong, the page is corrected, the review date changes, and where the correction changes the substance of what was said, the page states what changed.

We do not silently amend a figure that somebody may already have cited, and we do not treat a correction as an embarrassment to be minimized — a desk that publishes its files should expect to be corrected from them.

Published files are refreshed rather than versioned. A file at a given URL is the current computation, and there is no archive of superseded editions today. That limitation is stated on the download page rather than glossed, and it is the reason the citation format there puts the download date first.

Regulatory pages are re-retrieved on events, not on a calendar. California’s ADU chapter has been amended by six bills in under three years, including four chaptered on a single day.

A statute page here is re-run when a bill touching the chapter is chaptered, and the change is recorded in the change log with the section-level enactment lines rather than summarized.

Every page carries the date its evidence was gathered, at the top, where it cannot be missed. The "data as of" line is not boilerplate: on several pages it names different dates for different components, because the statute was retrieved on one date and the municipal harvest ran on another.

A reader deciding whether to rely on a page needs the older of the two dates, so both are printed.

If you find an error, the route is the contact details on the editorial policy page. A discrepancy someone found by recomputing a published file is more valuable to this desk than a figure nobody ever checked, and it will be treated that way.

What this method does not give you

The HyreADU method does not make the site a substitute for professional advice. Nothing here is legal, tax, financial, appraisal or construction advice. Zoning interpretation, underwriting and appraisal practice vary by jurisdiction, lender and appraiser, and the useful next step on a specific lot is the planning counter and a licensed professional.

It does not make the figures universal. A well-computed median from five jurisdictions is a well-computed median from five jurisdictions.

The method protects against a figure being wrong; it does not make a figure apply to a place it was not computed from, and the geography label on every store-based number exists to stop exactly that borrowing.

It does not resolve legal questions. Whether a particular local ordinance complies with state law is a finding for the Department of Housing and Community Development under Government Code § 66326, and a preemption conclusion belongs to a court. This site records what documents say. It names no jurisdiction as non-compliant.

And it does not eliminate judgment. Choosing which frame to compute over, which contrast to publish, and which interpretation is most defensible are all judgements. The response to that is not to pretend they are not judgements; it is to publish the file so somebody else can make different ones from the same rows.