Meeting 11/11/25

Weekly Meetings, Fall 2025, 2025

Helpful notebooks:

Potential Ideas:

  • What routes can a reciever can get the most seperation on dependant on field location (Redzone vs own 5 yrd line)
    • What about seperation?
    • Type of route compared to seperation
      • Focus on handful of routes
  • What defender is closest to the targeted reciever, probability of defender changing the play depending on the position they play
    • How does the probability change while the ball is in the air / as they move
    • What defenders are the best at affecting catch probability
    • If he didn’t change the play, was there a more optimal path (take into account reorientation)
  • How far does a reciever need to be from a defender for a reciever 100% not be touched by the defender
    • How to quantify being touched
      • Find / calculate arm length to create a bubble around each player
      • Does the play end within 5 yards of reception?
  • Who was able to put themselves in the right position
    • What is the right position
    • What were they doing
    • Are they in the best spot pre-snap to make a play
  • How quickly can defense correct
    • Switching direction
    • Coming back to the ball if they are too deep
    • Using the orientation and speed data
    • Is a saftey switch direction before the ball is thrown to go the right way
    • Data comes from chips in their shoulder pads, be careful using orientation with QB’s

Other Notes:

  • Create some visual with each step, not everyone reading your submission with come from a technical background
  • Take a “Who is the best corner at this specific thing?” type question and take it a step further
  • Use public data with the provided data to help
  • Don’t end analysis with a list of players that are the best at __
    • Go deeper with its application
    • Go into an individual play, what is systemic, what is a deviation, what is random …
  • Use the output data in your final metrics / analysis
  • Show what your model is doing on a play, like something that a TV broadcaster could use on the air
  • First good step is to reorient the data so it is all going the same way
  • Pick one play and build out the response for that one play, then apply it to the dataset