SportCrow's projections, rankings and start/sit calls all come from our own projection models, run by our own software. This page explains what goes into it, when it updates, how we check it, and where it falls short.
Where the data comes from
NFL. Schedules, betting lines, rosters, depth charts, injury reports, player and team statistics, snap counts and play-by-play come from nflverse, the open NFL data project. Team news about player status (who's out, who's back) comes from Big Balls, a sports data provider. We store the status and when it changed, not the news text.
NBA. Schedules, rosters, box scores, injury designations and betting lines come from Big Balls. Box scores from 2019 on are the record of playing time and production the basketball model learns from.
How a projection is made
Workload first. In football, the model estimates how many plays, passes and carries each team will have, then shares them out among its players by recent usage and depth-chart role. When a teammate is out, his share moves to the players who are left.
Efficiency turns opportunities into points. Catch rates, yards per touch and touchdown rates come from each player's history, pulled toward league averages where his sample is small. One big game doesn't count as much as a steady workload.
Basketball starts with minutes. The NBA model estimates each player's minutes, then his production per minute, and keeps every team's players adding up to a believable team total. Its minutes model learns from recent playing time, role, teammates' absences and rest.
The matchup enters through the game. NFL projections use betting lines and team tendencies, plus a learned correction from pre-game signals that include how the opponent has done against the position. NBA projections adjust for opponent strength, home court and betting totals where they exist.
Chance to play is part of the number. NFL availability comes from how players with the same injury designation and practice status have actually played in past seasons. NBA availability combines a learned chance of playing with the player's designation. Neither is a promise that a player suits up.
Then we simulate. The model plays each game out 10,000 times. NFL simulations share one game environment across both teams, so a shootout lifts both offenses together; NBA simulations vary minutes, pace and shooting. Every simulated stat line is scored in each fantasy format.
The projection you see is the median: the middle of the simulated scores. It already includes the chance that a player doesn't play, so a questionable star can rank below a healthy starter with a lower ceiling.
When projections update
Our data and projection jobs run every 20 minutes. New numbers appear when a provider has published something new, such as an injury report, a depth chart change or a line move.
At kickoff (NFL) or tip-off (NBA) a game's projections stop changing. That frozen number is what our accuracy reports grade, so we can't quietly improve a projection after the game has started.
How start/sit calls work
A start/sit call recommends the player more likely to outscore the other in your scoring format. We compare the two players' full simulated score distributions, not just their medians, so a steady player and a boom-or-bust player are weighed as they really are. The percentage is that estimated chance.
Player pages give a single-player answer instead ("Start", "Flex play", "Sit"), from where his median ranks at his position, sized for a 12-team league. None of these calls know your roster, so treat them as one input, not an order.
Rankings and tiers
Rankings order players at each position by median projection. Tiers break the list where the gap between two neighbouring players is unusually large, so a tier groups players with no big drop between neighbours.
How we measure accuracy
After every game we compare eligible frozen pre-game projections with what players actually scored, once the box scores are in. A projected player who doesn't record a stat line counts as zero. The reports use PPR scoring for the NFL and points-league scoring for the NBA.
The main number is the typical miss: the average distance between our median and the real score. We also check how often the projection landed closer than the simplest guess, the player's average so far this season, and how often it ordered two players correctly. The results are public: NFL accuracy and NBA accuracy.
Limits
- Not every rostered player gets a projection. In the NFL a player needs a place on the depth chart; in the NBA, recent minutes or a known return from injury.
- Injury news can land after our last run before a game. Check the final inactives yourself.
- Our historical NBA data has gaps (some games, older injury reports and betting lines), so tests on past seasons can't recreate everything a live projection knows.
- A projection is a range of outcomes summed up as one number. Low-probability games happen every week.
Report a mistake
If something looks wrong, use the Report a problem link on any page, or email hello@scorelineventures.com with the subject "SportCrow correction". Include the page, the week or game day, and what you think is wrong.
See also our About page, NFL accuracy report and NBA accuracy report.