Know the odds before the vote.
Forecasts of which California bills become law, for the people who write, track and influence them.
- 38,333 bills since 2007-08
- Tested blind on 2025-26
- AUC 0.70–0.95 as a bill moves
- Loading bills…
Loading the featured bill…
Watchlist
Select a bill to chart it.
Loading…
The odds sharpen as a bill moves
AUC: how often a signed bill ranks above one that failed (0.50 is chance). One model per checkpoint, trained on earlier sessions and tested on . It ranks bills; it is not the chance any single forecast is right.
Predicted against actual
Describe a bill. See its odds. Get a version that can pass.
Type a bill idea or paste a Legislative Counsel's Digest. A model trained on every California bill since scores its chance of becoming law. AI explains the number in plain English and drafts a stronger version.
In plain English
Written by AI from the model's numbers. The odds themselves come from the statistical model.
Describe a bill above and choose Score and explain it. You'll get why it scores where it does, where bills like it stall, and what would raise the odds.
Rewrite it to pass
AI drafts amended versions aimed at the weakest steps, like a staffer before a hearing. The model scores every draft and only shows ones that raise the odds, so the gain is measured, not claimed.
Model inputsAI fills these in from your description. Change any of them to see the odds move.
Where it could stall
Chance of clearing each step, for a bill that reached it.
What moves the estimate
Each line is the change in the estimate from that input, against a typical value: no flag, no coauthors, an average committee and filing window, and the party average for the author.
These are what the model has learned goes with success. They are not causal effects: a bill with coauthors may do better because it was already a consensus bill.
Similar past bills
Past bills whose titles share words with the subject.
Committee and author
Loading rosters…
How well the model is calibrated
Bills grouped by predicted chance, against the share that were actually signed (five-fold cross-validation).
Method and limits
Trained on earlier sessions and tested on , a session it had not seen, the model ranked a signed bill above an unsigned one about of the time (AUC , where 0.5 is chance). The range shows how much the estimate moves with the data it learned from and between typical sessions. It cannot see a bill's merits, the Governor's priorities, opposition, or deals made along the way.