What this is, and what it is not.
A personal forecasting project: I publish my probability estimates for events in Formula 1 and politics, log every revision, and let the record judge me.
The premise
Every forecast here is one person's subjective estimate. There is no crowd, no order book and no money — which means no price to hide behind. If a number is wrong, it is wrong because I reasoned badly.
The point is calibration. Publishing forces me to commit to a number before I know the answer, and the revision ledger makes it impossible to quietly rewrite what I thought last month.
How to read a forecast
The bar at the top of each page is the whole distribution: outcomes are mutually exclusive and always sum to 100%. A number moving up means another must come down.
The chart shows how each estimate has travelled. Numbered pins along its baseline mark revisions — click one to jump to the note explaining that move.
The percentage change beside each outcome compares against my previous published estimate, not against any market.
Formula 1
Built from race pace rather than qualifying, corrected for fuel load and track type, then adjusted for reliability record and remaining calendar. Single-race evidence moves my numbers slowly; repeated operational patterns move them faster.
Politics
Polling, fundamentals, institutional constraints and historical base rates, weighted by how far out the event is. Far from an election I forecast parties rather than people, because person-level estimates before a primary field exists mostly measure name recognition.
How often numbers change
It varies from minutes to weeks depending on the forecast. A championship market can move twice in a race weekend; an election three years out might sit still for a month. Anything that moves gets a dated note saying why.