Forecasts, on the record
Intellivoir forecasts under a protocol before it forecasts with a model: a resolvable question, a frozen cutoff, a named reference class, a card recorded before its outcome exists, resolution under a rule fixed at issue, and a score. The first populations are the service’s own state bills, whose histories it reads twice a day and whose outcomes resolve themselves. Every number is a pooled base rate with its counts beside it. Experimental and informational only.
How a card is made
- 1A resolvable question
Every card asks one thing an official record will settle: will this bill be enacted within 120 days, will it record another action within 30. No card asks whether a law is good or whether a trend continues.
- 2A cutoff
The card is conditioned only on what the service had read at the moment it was issued — the bill's status and last action date as stored then, and the corpus's histories as of the last completed sweep. Nothing learned later is allowed back in.
- 3A reference class
The forecast is a base rate: among bills at the same stage, quiet for the same number of days, how many had the event within the horizon? The state's own class is pooled toward the nation with a stated prior strength; the counts are on the card.
- 4Abstention
Below twenty fully observed comparable cases the card abstains and says why. An abstention stays in the denominator; the service cannot improve its record by staying quiet on hard cases.
- 5A ledger
The card is written before the outcome exists, hashed, and never edited. A bill that moves gets a new card; the old one stands as issued.
- 6Resolution and score
When the record answers, the card resolves under the rule written on it, with seven days allowed for publication lag, and is scored (Brier) against the national baseline. The record on this page is that score, and it will be thin for months.
The populations, and the record so far
Reading the scoreboard…
Open cards 0
Reading the ledger…
The engine is UPFF, a governed forecasting protocol whose reviewed implementation lives in the policy service. The populations here are the first it can honestly run; new ones — a data-centre item’s fate on a local agenda, a moratorium’s adoption, a permit’s issue — are definitions in the same shape. Public ledger: /api/ai-policy/forecasts/live. Not legal or investment advice.