Qipudata
San Francisco · v0.7.1

Known limits

Ordered by how much they can mislead you. None of them is hidden: each has a column or flag you can filter on.

Things that can change what a decision seems to say

  • First readings. A bill "passed for printing" is the first of two votes, not a law. In the v0.6 audit, 27 of 44 wrong outcomes were first readings shown as passed. Check the page before you write that something became law.
  • Ordinance and resolution numbers before 1998, in rows from the older reading passes: about three in ten are wrong, often the number of the neighbouring item. Look the number up on the page before you cite it.
  • Summaries. The one-line summaries are written by a model. About one in eight says more than the page, gets a detail wrong, or describes the wrong item. The quote and the page are the record.

Things that can put a decision in the wrong place

  • Street-only places. When the minutes say "Haight Street" with no number, we can only link the whole street. Most of those links in the audit sample were too vague or the wrong street.
  • Corners land on the right block about half the time.
  • Natural and large features (creeks, shipyards, freeways, landfill) are drawn as a circle around one point and marked approximate.

Things that can throw off a count

  • Rows are not decisions. 539,192 rows stand for 539,192 decisions: an item can be read by more than one pass. Count decision_group to count each item once; the grouping is careful rather than complete, so counts come out a little high. Each passage through the Board (referred, passed for printing, finally passed) is still its own decision.
  • Undated rows. 31,492 of 539,192 rows have no date in v0.7.1; year is filled for all of them.
  • Roll calls across a page break can lose names.

What isn't here yet

  • No person has checked a random sample (see Accuracy).
  • Summaries after 1951 are hidden on this site for now (see Privacy).
  • This release covers San Francisco. Vancouver tables ship separately.