What Are Insights?
Insights is a pattern discovery tool, not a prediction engine. It answers one question: how has this player performed in similar situations before?
This is an important distinction. Predictions look forward. Insights look backward. They show you averages, medians, and hit rates for a player under specific historical conditions — giving you context that raw box scores alone cannot provide.
Context Filters
Insights become powerful when you apply context filters. These narrow the historical data to situations that resemble the upcoming game:
- Pace — How fast or slow was the game? A player's rebounds in high-pace games may look very different from slow-paced games.
- Location — Home vs. away splits can reveal meaningful patterns, especially for players with strong home-court tendencies.
- Minutes — Filter by minutes played to isolate performances where the player had a similar role/workload.
- Opponent type — Was it a top-10 defense or a bottom-5? Opponent strength significantly impacts stat lines.
What You See
For any filtered set of games, Insights shows you:
- Average — The mean value across matching games.
- Median — The midpoint value, less affected by outlier games.
- Hit rate — What percentage of the time the player went over or under a given line in those situations.
- Sample size — How many games matched your filter criteria. More games = more reliable patterns.
How to Use Insights Alongside Predictions
Insights and predictions serve complementary roles. Predictions tell you what the model expects to happen. Insights tell you what has happened before in similar spots.
When both agree — for example, the model projects 8.5 rebounds and Insights shows the player averages 9.1 rebounds in similar pace/opponent contexts — that is a convergence signal. When they disagree, it is worth investigating why.
Common Mistakes
- Small sample sizes — A hit rate of 80% across 5 games is not reliable. Look for 15+ games before drawing conclusions.
- Over-filtering — Stacking too many filters at once can reduce your sample to a handful of games, making patterns meaningless.
- Confusing correlation with prediction — Just because a player averaged 25 points in past home games against weak defenses does not mean it will happen tonight.