Tracking Project Debrief & Plan Going Forward

I spent a lot of the time in the lead up to the season tracking old games and getting the behind the scenes structures prepped for a full season of tracked games. I would love to say it went well but it was at-large a frustrating and slow start experience.

I was tracking some games at the beginning of last season and was being a bit maniacal about the volume of tracked events. To make up for that, I wasn’t really tracking anything of the opposition and I ultimately didn’t continue for two reasons: 1. It was taking too long 2. ESPN starting deleting games from the archive.

In order to have tape to look back on and analyze, I made the decision to go into video breakdowns as a way to store and build some way to do specific player analysis. Really, I wanted to move forward into being able to tell the tactical stories of the games. I enjoy getting into the weeds and in the details and I think there’s a lot missing from regular day-to-day hockey coverage

If a player struggled, I wanted to show why. Hockey is so wicked and complex and I think moving towards understanding the team and matchup contexts would help explain certain player performances.

At the end of last season I made some improvements to the video review process but still felt like I was a little biased towards my preferred playstyle and preferences. I thought getting even grittier and recording these stats would fold back into the video analysis and build confidence in getting closer to objectivity. That these stats would also serve as raising flags for great analysis. If the Blue Jackets normally exit the zone with 80% success rate but this game it was 50%, why?

The problem: Time.

Tracking Project Debrief

Tracking to the depth that I desired, which was quite deep, took me around 9 full hours for each game(!). Some of that time was spent in deliberation of how to track a certain event, the painful hours smashing my head against these particular walls did yield plenty of insight into NHL hockey, some was in searching for the jersey number of the player that did it and some was still getting the tracked inputs automatic without having to spend much time recording them.

I am sure that I would have gotten that time down dramatically with practice but even in doing so the time still posed a difficult issue. Tracking games for the purpose of creating better game-to-game breakdowns is difficult to imagine being completed on a time scale that’s useful to anyone reading them without dramatically altering my sleep schedule.

I thought, and still think, that the work was worth doing, and the corresponding data based conclusions and visualizations that could be created in the 2025 offseason very exciting, but would probably result in a weekly “analytics” or “tracked data” recap. The problem, then, was that it moves my work into nearly exclusively data and away from the video work that I think is more useful in bridging the gap between stats and understanding for the greater public.

I wrote up an entire “White Page” for each metric and it’s intention within the tracking project. I call it a white page because I included all of the latest research I could find explaining the validity and importance of each metric. I created rigid definitions that I would use to create a theoretically repeatable process but continually tweaked them as I confronted the reality of the NHL.

Largely, the metrics would have looked a lot like Corey Sznajder’s tracked data, though our ultimate techniques may have differed, except I also intended on tracking some extra “possession improving” plays that weren’t discretely tracked namely off-wall plays but also including certain quick-ups and ice-stretching plays. Including this information was largely inspired by the work of Mitch Brown and Lassi Alanen.

If anyone would like access to the two full games I have tracked, I am more than happy to share them and explain the metrics and some potential ideas for creating more interesting information from them.

I was still somewhat committed to this approach but getting back into video breakdowns and looking for systems during the preseason was really fun and informative. Without a serious platform or people doing things with the data, I would kind of just be creating a cool but very specific dataset and continuing to exist in the relative shadows.

I’m ultimately serving two masters in terms of next day video reviews and hand-tracking and I’m not sure I can do both without dedicating an amount of time that would be highly disruptive to my actual money paying job and also the people around me. The first is telling the game-to-game stories about how and why (and also who) drove a win or a loss and the second is in bigger picture information about the ability and trajectory of specific players.

Going Forward

Going forward, I believe I’ve come to a conclusion and a method that will help serve my overarching objectives without drowning me in work. I’ll be tracking shots, shot assists (with corresponding levels of danger) and applying context to the nature of the shots such as the nature of their origin.

If done well I should be able to associate these numbers with an xG model and not only provide shot and chance contributions but also create xG, xA1 and xP1 metrics.

From the game-to-game perspective, applying context to these shots can help us gauge relative danger or style in a way that simply numbers could not. Relative contextual factors like: Rush vs In-Zone vs Forecheck vs Faceoff shots and chances. Odd-man breakaway vs forecheck turnover.

For the most part, shot differentials within matchups are still the biggest driving factors in wins and losses and utilizing the metrics provided by HockeyViz just seems like a no-brainer.

Applying situational context could just tell us different stories about the strengths and weaknesses of stylistic matchups against teams and also allow me to dial in a more “objective” film analysis. “The Blue Jackets gave up 3 high danger chances from forechecking turnovers, here’s what it looked like” style.

Without specific transition and/or forecheck data we’ll be a little blind for estimating certain types of contributions and how they might translate to future games. Having to-the-second transition/turnover, shooting and puck recovery information could have yielded spectacular data with respect to matchup/teammate/system contexts but I have to accept that if I want to write about hockey this season, I’ll need to make it take less time.

From here, hopefully I can continue to find systems matchups and player performances worth spotlighting while also creating a useful dataset for analyzing and understanding Blue Jackets players and their opposition within game context. Unfortunately, it will likely be difficult to properly evaluate the play of defensemen with this highly limited scope and that is a source of great frustration for me.

I think the ideal implementation of deeper tracked stats and film will be to cover elements of the game that aren’t well represented via counting stats and can better capture contributions in the build-up especially those of defensemen. I think Mitch Brown with Build-up/60 is on the closest track but improving on any of these metrics will be a project for another day it seems.