BRENNEN

Industries / Media, arts and recreation / Professional sports

Scouting and analytics

Scouting and analytics is how teams find, judge and sign players, and how they turn records of play, from hand-kept score tables to cameras that follow every player, into decisions about who plays and how.

6

Milestones achieved, from 1859 to 2016.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1859

    The box score appears

    Henry Chadwick, a sportswriter, printed a table for an 1859 game in South Brooklyn that listed every player and what each did at bat and in the field, turning a game into numbers that could be added up. The Baseball Hall of Fame says he then printed the first season totals for teams and players in 1860. The Hall credits him with inventing the batting average, but a SABR article says he "was not directly responsible" for it.

    baseballhall.orgsabr.orgsabr.org

    Credit

    Henry Chadwick

  2. 1947

    A statistician on staff

    Branch Rickey of the Brooklyn Dodgers hired Allan Roth, who started on Opening Day 1947 and stayed until 1964. Roth logged nearly every pitch, the ball-strike count, where each batted ball went, and how hitters did against left- and right-handed pitchers, and the club used his numbers for trades and the batting order. SABR calls him the first full-time statistician employed by a major-league team.

    sabr.org

    Credit

    Allan Roth and Branch Rickey

  3. 1950

    Counting every pass

    In 1950 Charles Reep began writing down every pass in the football matches he watched, and over the years built a record of tens of thousands of passing moves; in the 1950s he worked as a performance analyst with Wolverhampton Wanderers. He used his counts to argue for direct play, moving the ball forward in few passes. A 2005 study by Hughes and Franks allowed for how rare long moves are and found they produced more goals, which writers such as Jonathan Wilson took as a flaw in Reep's reasoning; the statistician Richard Pollard replied in 2019 that a move's length cannot stand in for a team's style at all.

    Lessoncount how often each kind of move happens before reading anything into where goals come from, and check that what is counted measures the style in question.

    opensportssciencesjournal.com

    Credit

    Charles Reep, with the statistician Bernard Benjamin, his co-author in 1968

  4. 2007

    Cameras track every pitch

    In November 2006 MLB Advanced Media, the league's digital arm, signed a six-year deal to put Sportvision's PITCHf/x system, which tracks the speed, movement and location of every pitch, in all 30 ballparks, and it was turned on league-wide in 2007. By 2015 Statcast, which combines TrackMan radar for the ball with optical tracking supported by ChyronHego, was installed in all 30 parks to follow the ball and every player. In 2020 Hawk-Eye Innovations put 12 high-frame-rate cameras in each park, which also track body positions 30 times a second.

    sportsvideo.orgsabr.orgsportsvideo.orgsportsvideo.org

    Credit

    Sportvision; MLB Advanced Media; TrackMan; ChyronHego; Hawk-Eye Innovations

  5. 2013

    Cameras in every arena

    From the 2013–14 season every NBA arena got a six-camera SportVU player-tracking system from STATS LLC; Engadget reported that 15 teams had already paid for it themselves. The league's app, NBA.com and NBA TV were to use the tracking data.

    engadget.com

    Credit

    STATS LLC and the NBA

  6. 2016

    A scouting database breached

    In July 2016 Christopher Correa, a former St. Louis Cardinals director of baseball development, was sentenced to 46 months in federal prison. In 2013 and 2014 he had broken into Ground Control, the Houston Astros' private system for scouting reports, draft rankings and notes on trade talks, using a variation of an old password an Astros employee had used when he worked for the Cardinals.

    Lessona club's scouting files are a trade secret worth stealing, so they need the same password and access controls as any other business system.

    publicradioeast.org

    Credit

    Christopher Correa (offender); sentenced by US District Judge Lynn Hughes

Watch for

  1. TacticAI moves to open play

    At the Google for Brasil event on 10 June 2026, Google DeepMind and the Brazilian club Palmeiras said Palmeiras is the first club to use TacticAI for open play; until then the tool covered only corner kicks. They said it can forecast play up to eight seconds ahead. Reports say the club's data science department uses it for analysis and describe use during competitive matches as a possibility, and no paper or accuracy figures for the open-play version have been published.

    Achieved whena peer-reviewed paper or an outside test publishes how accurate the open-play forecasts are, or the club or an outside reporter documents its use in competitive matches.

    thenextweb.comolhardigital.com.br

    Claimed by

    Google DeepMind and Palmeiras, June 2026

  2. AI analyst for 48 teams

    FIFA and Lenovo built Football AI Pro, an AI assistant that answers coaches' questions from FIFA's own match data before and after matches, not during play. They announced it in January 2026, and on 17 July 2026 said all 48 World Cup teams used it; that report quotes only FIFA's Arsène Wenger and Lenovo's Ken Wong, and no team, coach or outside party.

    Achieved whenteams or an independent party confirm on the record how teams used it in their preparation.

    sportsvideo.orgitwire.com

    Claimed by

    FIFA and Lenovo, January and July 2026

Open

  1. 01

    Event data for African leagues

    A public dataset of every pass, shot, tackle and other ball action, with team, time and pitch location, for every match of one season of an African domestic league (men's or women's senior). Most of it is detected by software from match video and then checked by hand, and it is published in SPADL, a standard layout for on-ball actions that the kloppy library can read.

    The brief
    Credit · unclaimed
  2. 02

    Open full-match tracking from broadcast

    An open pipeline that turns a full 90-minute broadcast into each visible player's position on the pitch in meters, with team and shirt number, and a public accuracy test.

    The brief
    Credit · unclaimed

Open 01 · the brief

Event data for African leagues

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public dataset of every pass, shot, tackle and other ball action, with team, time and pitch location, for every match of one season of an African domestic league (men's or women's senior). Most of it is detected by software from match video and then checked by hand, and it is published in SPADL, a standard layout for on-ball actions that the kloppy library can read.

Why it is still open

Hudl StatsBomb's public competition list has one African entry, the 2023 Africa Cup of Nations, a national-team tournament, and no African domestic league; its newest release (27 May 2026) added 771 matches from five women's leagues in Europe and the US. Searches found no open event data for any African domestic league. The best open method for spotting ball actions in video scored 60.03 out of 100 in SoccerNet's 2025 challenge (an average precision that counts an action as found if it is timed within 1 second and given to the right team), measured on English league games, so software alone is not yet good enough.

github.comhudl.comarxiv.org

Built with what exists

SoccerNet's sn-teamspotting code (12 action types, each with the team that made it); socceraction to write SPADL; kloppy to load and convert data.

  1. First step

    Choose a league and ask its office in writing for permission to process match video and publish the data, then hand-code one full match as the reference.

  2. Done looks like

    Every match of one season published under an open license, with written permission from the footage rights holder. A report gives, for each action type, precision (the share of detected actions that are right) and recall (the share of real actions found), measured against at least 5 matches hand-coded by two people, with how often the two coders agreed.

Open 02 · the brief

Open full-match tracking from broadcast

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open pipeline that turns a full 90-minute broadcast into each visible player's position on the pitch in meters, with team and shirt number, and a public accuracy test.

Why it is still open

The open benchmark, SoccerNet-GSR, uses 200 clips of 30 seconds, and its best 2025 score was 63.90 GS-HOTA, up from a baseline of 29.01; GS-HOTA is a 0–100 score that rewards placing a player within about 5 m of the true spot with the right role, team and shirt number, and teams said shirt numbers were the main difficulty. Full-match tracking from broadcast exists only from companies' own systems: SkillCorner publishes 10 Australian A-League matches and says "around 97%" of player identities are right, with no independent check described, and SoccerNet-GAR shares Gradient Sports' tracking for all 64 matches of the 2022 World Cup, "automatically extracted from broadcast footage and manually refined by annotators", under a non-disclosure agreement.

arxiv.orgarxiv.orggithub.comhuggingface.co

Built with what exists

sn-gamestate, SoccerNet's open baseline; kloppy, which already reads SkillCorner tracking, to export a standard format; SkillCorner's open data as a reference for file format and frame rate (10 frames a second); SoccerNet-GAR's tracking, where its terms allow, as a reference to score against.

  1. First step

    Run the sn-gamestate baseline on the SoccerNet-GSR validation set and match its published score, then run it on one full half and log where it fails.

  2. Done looks like

    Code plus tracking for at least 10 full matches, using only footage the team has the right to process. GS-HOTA reported on the SoccerNet-GSR test set and on at least 20 hand-labeled 30-second clips from the released matches, plus the share of player-frames placed within 5 m with the correct identity.