Everything the Board decided, 1906–2026

What this table can and cannot tell you

Every row has been read and sorted — 160,757 of 160,757. That does not mean every row has a subject. A row without one is a row that is not a decision — a running footer, an index row, a line cut in half, a bare vote outcome, or what a member of the public said about a file. None of those is an act of the Board, and none of them has a subject to give.

■ a decision, with a subject 132,162 ■ not a decision 28,595

28,595 rows (17.8%) are not a decision at all. Scanning a century of printed minutes picks up running footers, index rows, lines cut in half by a page break, and bare vote outcomes — “Resolution No. 341-25 ADOPTED”, where the description of what was voted on lives in a different row. They are kept, so what was set aside can be audited, and they are marked so nobody sums them by accident.

Of the 132,162 rows that are decisions, 132,162 (100.0%) say what they are about. There is no remainder: every decision in the corpus now carries one. Most rows were sorted by their opening formula, which is shared by thousands of records at a time and settles the verb without settling the subject: “the sum of $1,250 be and the same is hereby set aside, appropriated and authorized to be expended” is an appropriation every time, and the money is for a sewer in one line and for litigation in the next.

The subject, when the line carries one, is further along: after for the purpose of, in payment to, for the construction of. That clause was pulled out and read on its own — never keyed to the opening formula, which those rows share with thousands that are already sorted — and it recovered 5,419 subjects. What it did not reach was read afterwards, a sentence-shape at a time. Most of that turned out not to be a decision with a missing subject at all — it was a record cut in half by a page break, a supervisor’s remark during debate, or the Board’s own standing rule printed on the agenda — and those are typed as what they are rather than given a subject they never had.

How much of each decade says what it is about

Decisions only — the bar for a decade is the decisions in it, and every one of them names a subject, so no decade carries a grey remainder. Every decade has been read end to end. Shown by decade rather than as one figure because the Board’s language changes completely across the century — a single number would be carried by whichever era is biggest and would hide a thin patch in another.

How precisely the record says where

Not a yes or a no. A record that prints its own assessor block and one that says “somewhere on Mission Street” are both located, and only one of them can carry a dot on a map. Merging them would be inventing a precision the Board never wrote down.

What the record is about

Out of the 132,162 decisions that say what they are about — not out of 160,757 rows. Bare vote outcomes are left out on purpose: “Resolution No. 341-25 ADOPTED” is filed under running the city, which is true of the row and is not a subject; counting those rows in doubles that topic and moves it up the chart. Every decision has at most one topic, so these add up.

The street being first is the record, not a sampling artefact. Grades, sidewalk widths, acceptances and closures are what a nineteenth-century board of supervisors mostly did, and it wrote them in the most repetitive language it had. Every decision is on this chart — there is no set held back for want of a subject. Where an appropriation’s line never said what the money was for, the fund it came out of did: Sewer Fund, Municipal Railway Construction, Hospital-Jail Completion. That is the City’s own word for the purpose, printed in the record, not a guess about it.

How each row was sorted

The classified_by column says which. It is on every row so you can filter to only what was read directly, and judge the rest for yourself.

record

a person read this row

3,510

form

a person read the sentence-shape this row shares with others. The Board wrote the same sentence thousands of times with the address swapped; the shape was read once

93,823

purpose-clause

a person read the purpose clause — what follows “for the purpose of”, “in payment to”, “for the construction of”. These rows already had an action from their opening formula; the clause is what named the subject, and it was read on its own so that a subject written here could not reach back and re-topic the thousands of rows sharing the same opening

5,419

near-duplicate

the same sentence as a row a person read, with different scan errors. The Journal reprints an ordinance at every reading and the two printings scan differently

1,738

rule

a pattern that agreed with a person 90–100% of the time on a blind sample. Only three families earned this — the rest scored between nothing and two-thirds and were thrown away

36,129

The columns

Counting rules

So that two people counting the same thing get the same answer.

1. One row is one thing the Board dealt with.
2. topic is single-valued, so no row is counted twice — but the topics sum to the rows that carry one, not to every row.
3. Do not count rows whose record_type is not `decision` into subject figures — they are page furniture, index rows, or a bare vote whose description lives in another row.
4. Say which denominator you used: all rows, decisions only, or classified rows only. All three are legitimate and they differ.