What a start probability means, and how to read it
A start probability is our estimate of how likely a player is to be in his team's starting lineup for a given game. If we say 70%, we mean that in situations like this one, a player like him starts about seven times out of ten.
How to read the number
- It is about starting, not playing. A player who comes in as a pinch-hitter or defensive replacement did not start.
- It is for one game. Two games on the same day, as in a doubleheader, get separate probabilities.
- It is not a guarantee. A 90% player will miss about one game in ten that looked like this one. A single miss does not make a forecast wrong; a pattern of misses does.
- It changes with news. The probable opposing pitcher, injuries, rest days and a posted lineup all move the number. Once the official lineup card is exchanged, the question is settled.
- The projected lineup is the nine most likely starters. The probability next to each name tells you how settled that spot is. A lineup of nine players at 95% is very different from one with three spots at 55%.
Is a 70% really 70%? Calibration
A set of probabilities is well calibrated if things we call 70% happen about 70% of the time, things we call 20% happen about 20% of the time, and so on. Calibration can only be checked over many forecasts, which is why we publish it as a chart rather than judging single games.
Here is how our first, deliberately simple baseline did on the 2025 season. Its start probability is just the share of a team's last 10 games against the same kind of starting pitcher (left- or right-handed) in which the player started. It was tested on every 2025 game using only earlier games.
| Baseline said | Player-games | Average forecast | Actually started |
|---|---|---|---|
| 0% | 14,575 | 0.0% | 9.0% |
| 1–20% | 13,242 | 14.5% | 22.9% |
| 21–40% | 10,302 | 34.7% | 35.8% |
| 41–60% | 9,116 | 55.0% | 50.4% |
| 61–80% | 10,974 | 75.5% | 69.8% |
| 81–99% | 9,007 | 90.0% | 85.1% |
| 100% | 16,827 | 100.0% | 93.9% |
The middle of the range is close, but the baseline is overconfident at the edges. Players it gave 0% still started 9.0% of the time, and players it gave 100% started 93.9% of the time. That is what you would expect from a rule that only counts recent starts: it cannot see call-ups, returns from injury, rest days or trades coming. A useful model has to fix exactly this, and we will publish its calibration the same way.
One number for the whole forecast: the Brier score
The Brier score is the average squared gap between the probability and what happened (1 if the player started, 0 if not). Lower is better: a perfect forecaster scores 0. On the same 2025 player-games, the baseline scored 0.153. A forecast that gave every candidate the same probability scored 0.250. See the methodology page for details.
Where to find our record
Our accuracy policy explains how live projections will be locked before each game and graded, with calibration charts like the one above and comparisons against simple baselines.
Figures: our own calculations. The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org".