qipu data›San Francisco›Datasets›decisions

San Francisco Board of Supervisors decisions, 1906–2026

Name: San Francisco Board of Supervisors decisions, 1906–2026
Creator: qipu
Published: 2026-10-21
License: https://creativecommons.org/publicdomain/zero/1.0/

decisionsrelease v0.7.1CC0 1.0

About

Every item the San Francisco Board of Supervisors acted on from 1906 to 2026, as printed in its Journal of Proceedings and minutes. One row per item; an item read more than once can have more than one row. Each row links to the scanned page on the Internet Archive it was read from.

Where it comes from: Journal of Proceedings and meeting minutes of the Board of Supervisors, scanned by the San Francisco Public Library and held by the Internet Archive; recent minutes from sfbos.org. 93% of rows link to their archive.org page.
How it was read: The scanned text was split into items by rules. A language model (Claude Haiku) read the pages and wrote the one-line summaries; every model-written line is marked in summary_by.
How far to trust it: The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority. Details
Licence: CC0 1.0: use it for anything, no permission needed. The scans and other sources keep their own rights.
Version: v0.7.1, published 21 October 2026. A published release is never changed; corrections come as a new release.

Technical details

Table: decisions, rows where city = 'sf'
Publisher: qipu
Source: SF Board of Supervisors, Journal of Proceedings (1906–1999) and meeting minutes (2000–2026), scanned by the San Francisco Public Library and held by the Internet Archive; recent minutes from sfbos.org
Temporal coverage: 1906–2026
Spatial coverage: San Francisco, California
Rows: 718,093
Columns: 52
Distinct source pages: 155,860
Rows citing an archive.org page: 669,173 (93.2%)
Page link: exact on 620,576 rows; may be one or two pages off on 7,804; not page-level on 89,713
Instrument number: printed on 75,250 rows
Identifier: qipu:sf/decision/<identifier>|<item>
Version: v0.7.1, 21 October 2026
Licence: CC0 1.0. One-line summaries written by a model are marked in summary_by.
Formats: Parquet, SQLite; one page per decision in the record viewer

Rows

By origin

origin	Rows	Share
v2 read page by page by decisions v2 (claude-haiku-5-5)	344,936	48.0%
bos_corpus items read from the Journal text by rules	161,015	22.4%
resweep items re-read by claude-haiku-5-5	112,079	15.6%
roll_call roll calls with no item text	55,692	7.8%
site map rows with no item behind them	22,050	3.1%
map_blocks decisions placed on blocks for qipu's map	21,592	3.0%
hand written by hand from the cited source	729	0.1%

By outcome

outcome_best	Rows	Share
passed	325,076	45.3%
(not stated)	153,884	21.4%
referred	80,871	11.3%
passed first reading	55,604	7.7%
pending	44,188	6.2%
continued	25,044	3.5%
unknown	12,825	1.8%
refused	10,054	1.4%
tabled	3,343	0.5%
amended	2,489	0.3%

Rows without an outcome are mostly items the Journal records without a vote: communications, referrals, hearings.

By decade

Decade	Rows	Share
1900s	21,665	3.0%
1910s	64,132	8.9%
1920s	79,874	11.1%
1930s	47,299	6.6%
1940s	46,213	6.4%
1950s	50,360	7.0%
1960s	36,860	5.1%
1970s	62,467	8.7%
1980s	60,704	8.5%
1990s	58,878	8.2%
2000s	94,560	13.2%
2010s	63,501	8.8%
2020s	31,580	4.4%

By author of the summary

summary_by	Rows	Share
claude-haiku-5-5 (model-written; check the scan)	490,605	68.3%
a model in the 2026-09 pipeline (model-written; check the scan)	169,707	23.6%
(not stated)	57,052	7.9%
qipu (hand-written from the cited source)	729	0.1%

Method and accuracy

Method

The Journal text (OCR of the Internet Archive scans) is split into items by rules, and titles are classified by rules with some model labels (run bos-corpus-2026-09). Items from the 1960s to the 2010s were re-read one at a time by claude-haiku-5-5 (runs resweep-d1960-haiku55-2026-10 and later). Decisions v2 (run sf-v21-haiku-1014, claude-haiku-5-5) read the Journal and minutes page by page for 1906–2016; where a v2 row and an earlier row describe the same decision, same_as links them. Roll calls are parsed by rules (20_rollcalls.py). Each item's page is mapped to the archive.org leaf number from the volume's scan data. Names of private people are withheld for 75 years.

Methodology · runs

Accuracy

The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority.

Decisions v2 (run sf-v21-haiku-1014) against a key written by Opus 5.5 on 40 San Francisco pages, 115 decisions. The key has no record of a hand check; a Wilson 95% interval on 0.95 is about 0.89–0.98. Fields from the earlier runs are not measured.

Field	Metric	Value
number	exact	1.00
votes (counts)	exact	0.99
votes	confirmed on the linked page image / checked (Wilson 95% 0.85-0.99)	0.97
money	confirmed on the linked page image / checked (Wilson 95% 0.85-0.99)	0.96
finds	recall	0.95
action	exact	0.95
number+action+votes	all exact	0.95
vote names	exact	0.95
mover	exact	0.93
topics	F1	0.90
places	confirmed on the linked page image / checked (Wilson 95% 0.70-0.97)	0.90
outcomes	confirmed on the linked page image / checked (Wilson 95% 0.73-0.94)	0.87
money	F1 (vs Opus key)	0.80
interest	Spearman	0.78
places	F1 (vs Opus key)	0.66
parties	F1 (vs Opus key)	0.61
dates	F1 (vs Opus key)	0.55
related	F1 (vs Opus key)	0.17

Known limitations

Rows from decisions v2 and from the earlier runs can describe the same decision: 77,845 earlier rows are linked to their v2 row by same_as. Count distinct decisions with care.
Decisions v2 reads the previous page as context: in a 20-claim scan spot check, 18 quotes were on the linked page and 2 on the page before (v0.6 re-points them).
Block placement: from v0.5 a decision keeps a block only if a street named in its own text touches it, or a named site is within 400 m; 7.7% of placements moved to a street, a site or the city.
Numbers printed in the text are found by pattern; OCR errors (1G2 for 162) are not parsed.
The private-name filter is a pattern: it can miss a lone surname and hides some phrases that begin with a first name.

Columns 52

Column	Type	Can be empty	Privacy	Public copy	Description
decision_id	VARCHAR	no	none	as is	`qipu:sf/decision/<item id>`: corpus items `<volume>\|<n>`, re-swept items `<volume>\|g<n>`, hand-built `hand:<slug>`, map/site-only rows `map:...`, `su:...`, `sd:...`. Values e.g. `qipu:sf/decision/journalofproceed30sanfrich\|1` Can be empty no · Privacy none · Public copy as is
item_id	VARCHAR	yes	none	as is	The item inside its volume: `<volume>\|<n>` (corpus), `<volume>\|g<n>` (re-sweep), or a hand/map/site key. Can be empty yes · Privacy none · Public copy as is
origin	VARCHAR	yes	none	as is	How the row came in: `bos_corpus` (rules over the Journal text), `resweep` (re-read by a model), `roll_call` (a roll call with no item text), `map_blocks`, `site`, `hand`, `v2`. Values `bos_corpus`, `hand`, `map_blocks`, `resweep`, `roll_call`, `site`, `v2` Can be empty yes · Privacy none · Public copy as is
year	BIGINT	yes	none	as is	The year of the meeting. Can be empty yes · Privacy none · Public copy as is
date	VARCHAR	yes	none	as is	The meeting date (YYYY-MM-DD) when known. Can be empty yes · Privacy none · Public copy as is
date_from	VARCHAR	yes	none	as is	Where the date comes from (v0.6.1): `record` (the pass's own record), `page: running head` / `page: page heading` / `page: meeting heading` (the Journal page's printed meeting date before the item), `page: heading of an earlier page` / `page: previous page heading` (the last printed meeting date before the page), `page: dated claims on the page`. Can be empty yes · Privacy none · Public copy as is
rollcall_id	VARCHAR	yes	none	as is	v0.6.1: the roll call of this decision's item (its last; it may be printed on the next page). Can be empty yes · Privacy none · Public copy as is
decision_group	VARCHAR	yes	none	as is	One id per printed item (v0.6.1): rows of different passes over the same item (v1 corpus, resweep, v2, map, site, roll call) share it; it is the decision_id of the group's first row by pass. Count distinct decision_group for "how many decisions". Can be empty yes · Privacy none · Public copy as is
kind	VARCHAR	yes	none	as is	resolution, ordinance, motion, bill, hearing, report ... as classified. Can be empty yes · Privacy none · Public copy as is
heading	VARCHAR	yes	free text	private people's names withheld under the embargo	The item heading as printed (OCR). Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
plain	VARCHAR	yes	free text	private people's names withheld under the embargo	The run's one-line plain summary (model-written; see summary_by). Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
record_type	VARCHAR	yes	none	as is	What the row records: decision, vote-outcome-only, report ... Values `cut-in-half`, `decision`, `departmental-work`, `index-row`, `page-furniture`, `public-comment`, `too-damaged`, `vote-outcome-only` Can be empty yes · Privacy none · Public copy as is
topic	VARCHAR	yes	none	as is	Topic label. Can be empty yes · Privacy none · Public copy as is
subtopic	VARCHAR	yes	none	as is	Finer topic label. Can be empty yes · Privacy none · Public copy as is
action	VARCHAR	yes	none	as is	The last action on the item (adopted, passed first reading, referred ...). Values `accepted`, `amended`, `appointed`, `appropriated`, `commended`, `granted`, `ordered`, `other`, `paid`, `refused`, `settled`, `voted-or-scheduled` Can be empty yes · Privacy none · Public copy as is
outcome	VARCHAR	yes	none	as is	The run's outcome (passed / refused / rejected / vetoed / pending / unknown). Values `amended`, `approved by voters`, `continued`, `passed`, `passed first reading`, `pending`, `referred`, `refused`, `rejected`, `rejected by voters`, `submitted_to_voters`, `tabled` ... Can be empty yes · Privacy none · Public copy as is
source_url	VARCHAR	yes	none	as is	The archive.org page of the item. Can be empty yes · Privacy none · Public copy as is
block	VARCHAR	yes	none	as is	Assessor block the item names, when it names one. Can be empty yes · Privacy none · Public copy as is
streets	VARCHAR	yes	free text	private people's names withheld under the embargo	Streets the item names ("\|"-separated). Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
amounts	VARCHAR	yes	none	as is	Dollar amounts printed in the item (text). Can be empty yes · Privacy none · Public copy as is
how_placed	VARCHAR	yes	none	as is	How the item was placed on the map (block, street, no place given ...). Values `between two streets`, `names its own block`, `no place given`, `street address`, `street corner`, `street only` Can be empty yes · Privacy none · Public copy as is
classified_by	VARCHAR	yes	none	as is	What classified it: rule, form, model. Values `form`, `fund`, `near-duplicate`, `purpose-clause`, `record`, `rule`, `title` Can be empty yes · Privacy none · Public copy as is
hocr_leaf	BIGINT	yes	none	as is	Page index in the OCR (before the leaf map). Can be empty yes · Privacy none · Public copy as is
page_exact	BOOLEAN	yes	none	as is	True when the page link was converted from the OCR leaf with the per-volume leaf map; false = may be a page or two off. Can be empty yes · Privacy none · Public copy as is
who	VARCHAR	yes	free text	private people's names withheld under the embargo	Who the item concerns, in words (model-written). Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
place_text	VARCHAR	yes	free text	private people's names withheld under the embargo	The place as written in the item. Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
run_id	VARCHAR	yes	none	as is	The run that read it (see `runs`). Values `resweep-d1960-haiku55-2026-10`, `resweep-d1970-haiku55-2026-10`, `resweep-d1980-haiku55-2026-10`, `resweep-d1990-haiku55-2026-10`, `resweep-d2000-haiku55-2026-10`, `resweep-d2010-haiku55-2026-10`, `sf-v21-haiku-1014`, `van-v21-haiku-1014` Can be empty yes · Privacy none · Public copy as is
address	VARCHAR	yes	free text	private people's names withheld under the embargo	The street address the item names, if any. Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
amount	BIGINT	yes	none	as is	The main dollar amount, as a number. Can be empty yes · Privacy none · Public copy as is
quote	VARCHAR	yes	raw text	left out under the embargo	Raw OCR text of the item. Not published for records under the embargo. Can be empty yes · Privacy raw text · Public copy left out under the embargo
site_plain	VARCHAR	yes	free text	private people's names withheld under the embargo	The map's corrected one-line summary. Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
site_outcome	VARCHAR	yes	none	as is	The map's outcome value. Values `approved by voters`, `passed`, `refused`, `rejected`, `rejected by voters`, `vetoed` Can be empty yes · Privacy none · Public copy as is
site_ids	VARCHAR	yes	none	as is	The map's row ids for this decision. Can be empty yes · Privacy none · Public copy as is
outcome_best	VARCHAR	yes	none	as is	passed / refused / rejected / vetoed / pending / unknown (the map's value first, then the run's). Values `amended`, `approved by voters`, `continued`, `passed`, `passed first reading`, `pending`, `referred`, `refused`, `rejected`, `rejected by voters`, `submitted_to_voters`, `tabled` ... Can be empty yes · Privacy none · Public copy as is
plain_best	VARCHAR	yes	free text	private people's names withheld under the embargo	The best one-line summary (the map's corrected line first). Can be empty yes · Privacy free text · Public copy private people's names withheld under the embargo
instrument_id	VARCHAR	yes	none	as is	Ordinance/resolution number printed in the item (see `instruments`). Can be empty yes · Privacy none · Public copy as is
outcome_from	VARCHAR	yes	none	as is	Where outcome_best came from (site, run, roll call ...). Can be empty yes · Privacy none · Public copy as is
same_as	VARCHAR	yes	none	as is	Audit: the id this row was merged from (decisions v2 attached to v1). Can be empty yes · Privacy none · Public copy as is
city	VARCHAR	no	none	as is	`sf` or `van`. Values `sf`, `van` Can be empty no · Privacy none · Public copy as is
privacy_flag	VARCHAR	yes	none	as is	`may_name_private_person_after_1950` on Board text from 1952 on (records up to 1951 are open: the one cutoff, qipu_data/cutoff.py). Values `may_name_private_person_after_1950` Can be empty yes · Privacy none · Public copy as is
summary_by	VARCHAR	yes	none	as is	Who wrote the plain line (model name, "model-written; check the scan"). Values `a model in the 2026-09 pipeline (model-written; check the scan)`, `claude-haiku-5-5 (model-written; check the scan)`, `qipu (hand-written from the cited source)` Can be empty yes · Privacy none · Public copy as is
status	VARCHAR	no	none	as is	What the Board did with this item at this reading, from the first rule that applies (status_from); never blank: `not_recorded` when nothing says (qipu_data/status.py, usability audit 19/10). Values `introduced`, `proposed`, `referred`, `continued`, `amended`, `hearing_held`, `first_reading`, `submitted_to_voters`, `recommended`, `claim_filed`, `demand_listed`, `reported` ... Can be empty no · Privacy none · Public copy as is
status_word	VARCHAR	no	none	as is	The status in the word people see (proposed, referred, first reading, passed, adopted, vetoed, failed, withdrawn, claimed, listed for payment, recommended, reported, no status on record ...). Can be empty no · Privacy none · Public copy as is
status_plain	VARCHAR	no	none	as is	The status in one plain sentence, e.g. "Sent to committee; no final vote found"; says when the matter went further later. Can be empty no · Privacy none · Public copy as is
happened	VARCHAR	no	none	as is	decided (finally passed, adopted, voters approved, veto overridden) / not_decided (proposed, referred, continued, first reading only, claim, recommendation) / failed (refused, tabled, withdrawn, vetoed) / unknown (no status on record). Values `decided`, `not_decided`, `failed`, `unknown` Can be empty no · Privacy none · Public copy as is
carried_out	VARCHAR	no	none	as is	Whether what was decided was carried out (paid, built, demolished): only with its own evidence (carried_out_evidence); unknown = not confirmed. A decision to pay is not a payment. Values `True`, `False`, `unknown` Can be empty no · Privacy none · Public copy as is
carried_out_evidence	VARCHAR	yes	none	as is	The statement or record that shows it was carried out. Can be empty yes · Privacy none · Public copy as is
status_from	VARCHAR	no	none	as is	Where the status comes from: the instrument lifecycle stage of this reading, outcome_best, the v2 extraction, another reading of the same printed item, the printed roll call, a wording rule on the heading/summary/printed text, or none. Values `instrument_lifecycle`, `outcome_best`, `x_decisions`, `other_reading`, `roll_call`, `wording_rule`, `none` Can be empty no · Privacy none · Public copy as is
matter_status	VARCHAR	no	none	as is	The furthest status of the whole matter (instrument lifecycle; else the same printed item; else the same title decided within 120 days; else this reading's own). Can be empty no · Privacy none · Public copy as is
matter_happened	VARCHAR	no	none	as is	happened, for the whole matter. Values `decided`, `not_decided`, `failed`, `unknown` Can be empty no · Privacy none · Public copy as is
matter_status_from	VARCHAR	no	none	as is	Where matter_status comes from. Values `instrument_lifecycle`, `printed_item`, `same_title_120d`, `this_reading` Can be empty no · Privacy none · Public copy as is
plain_fix	VARCHAR	yes	none	as is	Set when plain_best was rewritten because it stated an open matter as done (rule_* = wording rule, model_haiku = rewritten by Claude Haiku and checked); `plain` keeps the original. Can be empty yes · Privacy none · Public copy as is

Privacy: what kind of text the column holds. Public copy: what the published file does with it; the embargo covers records after 1951.

Preview

Date	Kind	Outcome	Summary	Printed page
8 November 1906	resolution	passed	Board tells Ocean Shore Railway to finish its stalled Potrero Avenue roadbed work at once, without further delay.	page ↗
8 January 1908	resolution	passed	Payment of $2,000 (≈ $74,876 in 2026 dollars) to Annie Galloway approved for a small lot added to the Jean Parker School site.	page ↗
8 July 1940	ordinance	referred	Terms for leasing City airport land by competitive bid were referred to the Finance Committee.	page ↗
23 September 1940	resolution	passed	Supervisor Arthur M. Brown, Jr. granted a leave of absence from September 28 to October 12, 1940, with permission to leave the State.	page ↗
11 March 1943	other	referred	Jack Smith requested higher pay for Rammer; Board moved to fix it at $11 (≈ $214 in 2026 dollars) per day and referred it to Civil Service.	page ↗
19 May 1943	other	passed	Board approved $3,686 (≈ $71,790 in 2026 dollars) for War Memorial overtime payments.	page ↗
26 May 1945	ordinance	not read	The annual salary ordinance fixing city positions and pay for the fiscal year ending June 30, 1946 was taken up for second reading.	page ↗
16 July 1945	motion	referred	Supervisor MacPhee's push for a new Juvenile Court Building was referred to the Public Health and Welfare Committee.	page ↗

Five of 52 columns, for eight rows from 1906–1951 sampled at build time. The date opens the record; every column is in the Columns list above and in SQL.

Download

Format	File	Size
Parquet	decisions.parquet ↗	89.2 MB
SQLite (all tables)	qipu_sf.db ↗	6.44 GB

SQL

SELECT year, date, plain_best, outcome_best, source_url FROM decisions WHERE city = 'sf' AND plain_best ILIKE '%Hunters Point%' ORDER BY date

Run in the SQL console (DuckDB in the browser, over the Parquet file).

Python

import duckdb duckdb.sql("SELECT * FROM 'https://tiles.qipu.org/data/releases/v0.7.1/parquet/decisions.parquet' LIMIT 5").show()

Download

File	Size
decisions.parquet ↗	89.2 MB

Cite as

qipu (2026). Board of Supervisors decisions (table: decisions), release v0.7.1 [data set]. 21 October 2026. https://data.qipu.org/usa/san-francisco/datasets/decisions/. CC0 1.0.

Cite a single decision

By its scanned page: SF Board of Supervisors, Journal of Proceedings, <year>, page n<leaf>, archive.org/details/<identifier>. The record viewer gives each decision its own citation.

Versions

Release	Date	Rows
v0.7.1	21 October 2026	718,093

Each release keeps its own folder of files (this one: releases/v0.7.1/), never changed after it is published. Some earlier releases have been withdrawn. Changelog

Related tables

Table	Rows
x_roll_calls	171,123
x_roll_call_votes	2,084,507
instruments	72,224
decision_numbers	88,876

Documentation

Data dictionary · Methodology · Accuracy