berkeleybuild.com · for reporters

Query Berkeley's housing record directly — no records request required

Everything on this site derives from one database, rebuilt from primary sources (every permit 2015–2026, audited row-by-row against the state filing). That database is public. The links below open it in your browser — a full SQL workbench, no installation, no account — with the question already typed in. Edit anything and re-run.

Start with a story: 2700 Shattuck

The ZAB approved the 359-unit mixed-use project this month (Daily Cal). Our record adds the clock behind it: SB 330 pre-application Nov 2023, full application May 2024, deemed complete Oct 10, 2024 — then 21 months under consideration, versus a 410-day median for major projects. And what approval means statistically: entitled major projects enter a waiting room currently holding 5,111 units, where 2,334 of them are already past the one-year lapse-eligibility line of BMC 23.404.060.

▶ 2700 Shattuck — every recorded event
The project's full timeline, straight from the database.

▶ The case-file notebook (Colab)
A guided walkthrough: the timeline, the stage clock vs the medians, the density-bonus/affordability context, and what happens to projects after approval — every number computed in front of you.

The standing questions, pre-typed

▶ The waiting room — every entitled project with no building permit, largest first. (33 projects, 5,111 units as of this writing.)

▶ Density-bonus projects — who is using the State Density Bonus (the law behind the 2700 Shattuck waiver debate).

▶ The stage clock — how long each major project took from application to entitlement, slowest first.

▶ Open the whole database — every table, write your own SQL. The main table is v_projects_flat; events are in project_events.

The documents — the developers' own PDFs

For a growing set of projects we archive the actual submissions — plan sets, resubmittals — as permanent PDFs (the city's own links expire; ours don't, and each file carries a checksum). Download them, then feed them to an AI notebook: NotebookLM is the easy path (upload the PDFs, ask questions, get cited answers), or run a model on your own laptop so nothing leaves your machine — see our guide to running open models locally, essential when a story involves sources you must protect.

▶ 2700 Shattuck — the submitted plans — all five archived PDFs for the story's project, with permanent links.

▶ Every archived document — the full shelf, project by project. Always verify an AI's claims against the record: the database is the ground truth, the model is a reading assistant.

Ground rules & citation

Cite as: "berkeleybuild.com, an independent reconstruction of Berkeley's housing records from city permit data" — and please link the page you used.
Every figure traces to a public record; the method and its audit live at the Audit page (our count vs the city's differs by 130 units of ~4,200, every row named).
Dates carry precision honestly: a few records only know their year (they appear as Jan-1). When something matters for publication, ask us and we'll pull the primary document.
"SQL query took too long"? The browser workbench caps queries at one second — on this 4 MB database that almost always means a missing JOIN condition (check your ON clauses, add a LIMIT). For genuinely heavy queries, run Datasette on your own machine with a higher limit: pip install datasette, download the database file below, then datasette berkeley.db --setting sql_time_limit_ms 3500. Or send us the query and we'll bake it into the database as an instant view.
The database updates when the record does (harvest + records requests); the file you query is the dated public mirror at /data/berkeley_housing_v2_public.db.

Questions, corrections, or a records-request you want cross-checked: via berkeleybuild.com. The whole pipeline is open — a reporter who wants to rebuild the record from scratch can: the curriculum.