Site sourcing and screening — California

What the two weeks actually look like

There is no phase of a deal called "data acquisition." Data acquisition is a standing tax levied on every other phase, paid by whoever is least senior enough to argue. In a real shop the pattern is consistent: an analyst has a site, and over roughly the next two weeks works through the following.

Roughly half of those steps involve reading a PDF and typing numbers into Excel by hand. The distribution is not even.

Two decisions come out of this phase. The first is whether to pursue the site at all. The second — usually within days of the first — is what price goes into the LOI, and that is where screening quietly becomes underwriting. Under 4 CCR Section 10325(c)(9)(A), if the paid purchase price exceeds appraised value, the leveraged soft resources amount in the tiebreaker is discounted by the overage, unless the Executive Director waives it under Section 10327(c)(6). A developer who overpaid for land can be scored out of the round for it. That consequence is rarely modeled at LOI time, and it is pure arithmetic if the tiebreaker is modeled before the price is agreed.

The parcel layer is a fleet of adapters, not a dataset

There is no single authoritative statewide California parcel layer with attributes. There are three partial answers, and the working one is per-county ArcGIS REST services. That is not a "write one adapter" problem — it is a fleet of fragile adapters that need monitoring, and it does not get easier with expansion. Every new county is a new adapter, and every new state is a new legal regime for whether assessor data is even public.

The breakage is real and current.

The fallbacks are worse than they sound.

Zoning: the code is a pointer, and the pointer has a vintage

Zoning normalization is usually hand-waved as "every city is different." The real reason is more specific: a zone code (R-3, MU-2, PD-14) is a pointer into a municipal code document, not a value. Two cities' R-3 are unrelated. The density, height, and parking that actually drive a feasibility model live in a table inside a PDF ordinance, and in overlays, specific plans, and planned developments that override the base zone. There are 539 jurisdictions in California, each amending independently.

For Southern California, SCAG's regional land use layers are the best available answer and are parcel-keyed.

Vintage is the weakness, and it is measurable.

Roughly a third of Riverside jurisdictions have zoning of unknown age. Any zoning-derived number must be displayed with its vintage attached, and where the vintage is null that must be shown as null rather than silently omitted.

The statewide fallback — California Statewide Zoning North/South, published by the Office of Land Use and Climate Innovation — is open, downloadable in unusually many formats, and older. Its own metadata is candid: data was collected from 535 of California's 539 jurisdictions, with PDF and image maps geo-referenced and transposed onto assessor parcel geometry, collection beginning in late 2021 and mostly finished in late 2022. The state disclaims accuracy outright.

One asymmetry belongs in the data model rather than papered over: SCAG covers five of the six counties in EZFeasi's footprint. San Diego is outside SCAG and needs SANDAG's own layer or the statewide fallback.

The hazard layers you can query are not the ones that kill deals

This is the most important structural fact about environmental screening, and it should be stated in the product rather than buried in terms of service. Flood and fire — the hazards that mostly cost money — are wide open.

The hazards that end deals are the ones behind a wall.

There is no engineering fix for the last category, and there should not be.

The corpus disagrees on EnviroStor specifically, and the disagreement is worth surfacing rather than resolving silently. One source lists EnviroStor as bot-blocked and puts it on a do-not-automate list; another lists it as verified-live and machine-readable. They tested different endpoints — the 403 came from envirostor.dtsc.ca.gov/public/api/site_search, while the ArcGIS Public Data Export responded and carries an apn field, meaning it joins directly to a parcel roster with no spatial work. The do-not-scrape-past-a-403 principle is correct; applying it to the dataset name rather than the endpoint suppressed a near-free build. GeoTracker really is blocked — link out, do not work around.

The practical consequence for anyone building or using a screen: absence of a hit on the layers you can query says nothing about the layers you cannot. A green checkmark across every published environmental layer implies a cleanliness the screen cannot support, and contamination discovered at Phase I or Phase II — after site control, with escrow money already spent — is a recurring deal-ender. Seismic layers carry a related trap: "not in a zone" and "never evaluated" are different answers, and a screen that collapses the unevaluated layer into a clean result is reporting something false.

Screening evidence is not certification

4 CCR Section 10325(c)(2)(N) makes "certification of site amenities, distances or service amenities that were, in the Executive Director's sole discretion, inaccurate or misleading" a negative-points offense. Negative points attach to people and firms, not projects — Section 10325(c)(2) allows up to 10 negative points per project and/or per violation, assignable to general partners, co-developers, management agents, consultants, guarantors, or any member or agent of the Development Team, across 23 enumerated grounds.

In a field like that, a single negative point is fatal for as long as it applies.

This is the largest liability surface in any automated screening product, and the discipline is simple: a computed distance is screening evidence with a cited source, to be independently verified by a human before it is certified. It is never the certification itself.

Two related corrections belong here because they are cases of a screen implying more than it knows. First, an Opportunity Area designation does not, in 2026, confer a threshold basis limit benefit in Southern California: the 10% High/Highest Resource basis-limit increase is gated on the county's unadjusted 9% two-bedroom limit being at or below $500,000.

Second, a transit tier derived purely from HQTA captures the Government Code Section 65912.123(b)(1)(E) density override and none of AB 2011's actual gating — treating it as an eligibility signal materially overstates AB 2011 availability.

Where the screen stops, and what it cannot tell you

The further right you go in a deal's life, the worse the data gets, which is the inverse of what a feasibility product wants.

Construction cost is the biggest hole in the landscape. There is no free, authoritative, California-specific, unit-level affordable-housing cost dataset.

Jurisdiction behavior is partially answerable and almost nobody uses it.

What APR data does not give you is entitlement duration: it tells you a permit issued in a year, not that the application took 31 months. Per-project entitlement duration does not exist as a public dataset anywhere in California. It lives in city permit portals with no public API, in staff memory, and in land-use counsel's heads. CEQAnet is not a substitute — ministerial pathways generate no CEQA document at all, so a CEQAnet-derived sample is systematically composed of slow discretionary deals, which is precisely the path a streamlined affordable project is trying to avoid.

One free asset worth wiring in at screening: CTCAC's List_of_Projects.xlsx.

It carries both APN and census tract, so it joins directly to a parcel table. It gives unit counts by bedroom and by AMI band, annual federal award, total state award, developer, general partners and management company.

Finally, the honest limit. Whether the seller will actually extend the option, whether the city planner will fight the project or fast-track it, whether a competitor is about to file in your region and at roughly what tiebreaker — none of that is a dataset, and none of it should be synthesized. Give the user a place to record their own judgment and its date. What screening can and should do is make sure the arithmetic questions are answered before the phone call: the region's available apportionment, the tiebreaker at the LOI price, the site control clock, and the hazard layers you can actually see.