Averaging vs conditioning
A pregame 60% is an average over thousands of imagined versions of the game: some blowouts, some comebacks, some close finishes. Once the game starts, most of those versions become impossible. The live number only averages over the versions still consistent with what has happened.
- Pregame: Team A 60%.
- A falls behind by 14 with 20 minutes left.
- Of all pregame scenarios, only the ones where A comes back from 14 down remain. Live A ≈ 25–35% in a typical model.
Neither number was "wrong". They answer different questions with different information.
Grading them fairly
- Grade pregame models only on pregame probabilities, locked before the start. PropBetEdge pick records do exactly this — a call locks before the event and is never rewritten.
- Do not judge a pregame model by the live chart. A team that led all game can still have been a pregame underdog, and that is fine.
- Fights are an extreme case: before the bell, a finish anywhere in the scheduled rounds is possible; after two rounds without one, the remaining distribution is very different.
Limitations
Comparing a live price with a pregame model probability is comparing different information sets. Only compare like with like.