How this works

One page per measure: what a YES and a NO vote do, why the measure is on the ballot now, the money involved, who placed it on the ballot, who endorses it, who is paying for each side, and the strongest case for and against. Seven measures also get a timeline of earlier decisions and a record of what past measures promised and delivered.

We start from primary sources: the official voter guides, the Controller’s and Legislative Analyst’s fiscal analyses, the legal text, campaign finance filings and published endorsements. An AI model drafts each page from those sources only, and every sentence carries a link to the source that supports it.

No person reviews or edits pages before they go live. Instead, every page must pass these automated checks. The last column shows how many published pages passed each one.

AI models from Anthropic write and check every page. This table is counted from the records each published page carries.

The cross-check by a model from another company was not run for this election, so it shows as skipped above.

We never endorse. We give each side’s strongest argument in its supporters’ terms. YES and NO use colors with no party or good-and-bad meaning, and both ovals on our pages stay empty.

The coalition scale runs from the progressive coalition to the moderate coalition. A measure’s position comes only from which tracked groups endorse or oppose it and who funds each side, using the group classifications listed below. It is not our opinion of the measure.

For state measures, the scale runs from the California Democratic Party and its allies to the California Republican Party and its allies, such as the Howard Jarvis Taxpayers Association. The position is set the same way: only by which tracked groups endorse or fund each side.

San Francisco campaign money comes from SF Ethics Commission filings on DataSF, which update daily. State measure money comes from the California Secretary of State (Cal-Access). Every money figure shows the date of the latest filing we include.

The cost examples on measure pages, such as a homeowner’s yearly bill, are calculated by code (ai/impact_costs.py) from the tax rates and amounts cited on each page. An AI model does not write or estimate these figures. Each example states its assumptions, such as building size and assessed value, so your own bill will differ.

Pages are translated into Chinese (Traditional and Simplified), Spanish, Filipino, Vietnamese and Russian. Terms are checked against the Department of Elections glossary, a third model translates back to English to catch changes in meaning, and wherever San Francisco or the state publishes an official translation of the ballot question, we show that exact text. There is no official Russian translation, so Russian pages must meet the strictest agreement threshold.

AI models write and check these pages. We label every page so you always know, and we show when each page was last checked.

Every section has a Report an error button. Every fix is listed in the public corrections log with the date and what changed.