Video summary
I Built A Tennis Simulator. It Thinks Forehands Don’t Matter.
Main summary
Key takeaways
Summary
The creator builds a tennis match simulator using ATP match data, then tests a hypothetical “Frankenstein” player against top opponents.
How the Simulator Works
- Players are represented by four ratings: serve, return, forehand, and backhand. Ratings are percentiles relative to eligible hard-court player seasons since 2000, with 50 representing roughly average.
- For each point, the model compares the server’s serve with the opponent’s return, then uses shot ratings to calculate the server’s chance of winning. It simulates points through games, sets, and matches.
- Each matchup is simulated 10,000 times to produce a more stable estimate of each player’s win rate.
- The model uses recorded tennis data and official scoring rules, but does not account for factors such as fatigue, morale, or tactics.
Key Finding About the Model
Testing revealed that changing a player’s forehand or backhand ratings did not affect the results. In this model, serve and return ratings effectively determine the outcome; the other two ratings have no influence. The creator suggests that serve-and-return results already capture much of what the other strokes contribute. As a result, the simulator’s initial four-stat model is less meaningful than intended.
Building and Testing “Frankie”
The creator allocates 150 rating points mainly to serve and return:
- Serve: 97, based on Jannik Sinner’s 2025 rating
- Return: 51, based on Taylor Fritz’s 2025 rating
- Forehand: 97, based on Carlos Alcaraz’s 2025 rating
- Backhand: 97, based on Alexander Zverev’s 2025 rating
Because the model ignores forehand and backhand, those ratings are effectively free. The creator also cautions that combining one player’s serve rating with another player’s return rating may not represent a real player, since the original metrics include broader player characteristics.
Match Results and Visualization
A 3D animation visualizes the matches. Fast lines represent most points, while set points and match points receive more detailed shot animations. The display also tracks match progress and win percentages.
Frankie wins the featured single match against Sinner, but wins only about 22–23% of matchups across 10,000 simulations, making Sinner the expected overall winner. Frankie loses to Alcaraz and loses more decisively to peak Djokovic.
The results differ from the creator’s predictions. The Sinner match also illustrates how an individual simulated match can differ from the overall win rate across 10,000 simulations.
Main Speaker and Sources
- Main speaker: The video’s creator and narrator from the AEJ channel
- Data source: ATP match and shot data covering seasons since 2000, as described in the video
- Coding tool: Claude, which helped create the simulator
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.