Many lifetimes ago, when I played Pee Wee football for the Baldwin Bombers, our head coach (Mr. Lamonde) would always talk about the “intangibles” of the game.* A kind of otherworldly sense a player has that you couldn’t really quantify (or even teach). I guess some 10 year olds shine, and some don’t 🙂
I’m still fascinated by this idea, despite being far, far removed from the game—which has changed dramatically in the 40 years since I suited up. I mean the financial stakes alone for some of these elite teenagers today is hard to wrap your head around. Luckily, those problems were never mine when it came to football. The game is dear to me given I played it during the formative years of elementary school, but once I joined the Junior High School team it became apparent I was mediocre at best.
I still enjoyed playing the game a lot, even if it was becoming more and more of a slog with a commitment of 5 or 6 days a week. By my sophomore year in high school I snapped my arm pretty bad skateboarding, and was in a cast for much of the rest of high school. So my paltry football career would end about where it should’ve 🙂
But back to the point at hand, I was not one of those players that found themselves in the right place at the right time to make the play: the oft heralded “big play makers.” But I love the whole idea of this sixth sports sense. As I’m playing Madden 2001 for the AI Maddeness season, part of the limitation is that the game does not preserve certain stats. For example, game-by-game stats don’t exist, which is why I’m manually capturing and scraping all the data for every player on the Jets week-by-week. Additionally, it only shows league leaders in passing, rushing, receiving, defense, etc., by certain parameters, so the top 25 receivers will be those with the most receptions, not necessarily those with the most yards.
It seems to follow that the more receptions a player has, the more yards they have, right? Not always, in the case of Wayne Chrebet, for example, he has 8 receptions for 227 yds and 3 TDs. These numbers are better than many of the league leaders (such as Tampa Bay’s Keyshawn Johnson) when it comes to reception yards, but because his receptions numbers are lower, he’s never featured in that category. This is just one example of how the stats for Madden 2001 paint an often incomplete picture. The possibility to sort by TDs, receptions, overall yards, or even reception percentage is not available, which is understandable given the designers of Madden 2001 were probably not aiming for comprehensive stats—or maybe it was a technical limit? I’m not sure, but I imagine in subsequent years they have developed more comprehensive stats that can be filtered by category—which could bode well for future years of this experiment 🙂Â

The Buccaneers Season Stats through Week 3
With incomplete player stats I’m beginning to think ChatGPT can actually do what it does best: hallucinate the details of simulated games each week. Because while Madden 2001 has an incomplete picture of player stats, it keeps a pretty comprehensive tally of a team’s season stats: things like total offense, defense, turnovers, third down conversions time of possession, etc. So, my experiment this week will be to get the comprehensive season stats for 2-4 teams that are playing each other prior to week 4’s game, and then compare them to the season stats after the week’s game is simulated. This should give me numbers in the aggregate that I can then ask ChatGPT to compare and then hallucinate something that never happened, albeit guided by the overall stats.

More Team Stats for the Tampa Bay Buccaneers going into week 3
I’m wondering if ChatGPT will be able to intuit the intangibles for certain players and the big plays that make the game magical. I mean, that’s kind of the thing about any sport, right, watching an individual or team move beyond the stats and provide a sense of wonder. That was the magic of the late 90s Yankees. It’s part of the mythos of sports, and even when I play this graphically challenged 2001 video game I find it captures some small sense of that—which I know sounds crazy. Maybe fake can be just as good?
____________________________________
*Another way to say it might be instinct.


What we call ‘intangibles’ are to a neural network patterns that don’t have a label (yet).
Speaking of intangibles. I had enrolled into engineering as transfer student to a big ol’ land grant University in Virginia. I had an advisor in the department that I had chosen to major in within Engineering. And that guy liked football metaphors. He tried very hard to let me down easy as I was struggle-bussing that first year. I think he said, words to the effect of, “I’d like to be a starter for a Major League team, but at my age that’s never going to happen”. The intangible here is when you’re not good at something, you have to give up. But me being somewhat “dense” maybe (or on a spectrum? I dunno), never got the underlying meaning of it until almost 20 years later. So some “intangibles” become tangible,… given time. Never Stop tangibling. ?
I wonder what the data looks like under the hood where the game stores it. Can it be accessed? What exactly gets stored in memory? What could be stored if we could intercept (pun?) it?
I listened to some crazy stuff on building a Sega game that makes me wonder. You might appreciate it (or one of the kids, if memory serves).
Funny you should mention this, I actually bug a bit into the game to see why so much is hidden, and it turns out I was wrong, all the data is there. There is a system where you can filter stats by league, conference, division, and team, so I can get everything—so this was me being stupid. It also brings up all sorts of opportunities, because if the game has all this data and it is emulated on a raspberry pi, like mine, what would it mean to just feed it right from the emulator to OpenAI specific GPT, now that would be sick. It would just dump the data there and it would automate the standings, full league stats that are actually searchable, players of the week, etc. Digging into the code of Madden 2001 would be a total blast, but I am definitely out of my league there (pun intended). The Sega gmae stuff has me inspired that maybe some genius out there has already figured this out.
The other piece here worth note is right now my emulator has a bug where sometimes the game freezes at the most inopportune times. During a play this is no big deal, it just forces a delay of game and then it resets and game play resumes. But if this happens while perusing the stats the game has to be reset and all the game stats are lost. I am basically starting that game from scratch even though I played the week’s game, it is very frustrating. I have been saving all the game stats via screenshots so I have the “official” stats, but the game system stats are different because I am forced to simulate the game until the Jets win to move on to the next week. That divergence creates extra work for me weekly to reconcile the numbers. This is a bye week, so i am planning on re-installing the Madden 2001 rom for PS1 in hopes that glitch is not native, but if it is th idea of regularly dumping data from the system to a file would be amazing. Also, the emulator (Batocera) has save states that I might be able to invoke, but if this bug hits while I am browsing stats during halftime I have forced a reboot, which would suck—although i could always just replay it. But every game is sacred and unique, so to lose it at halftime, even if I was down, would be against the spirit of any given Sunday. Anyway, I am far afield from you comment, but I have been meaning to get to this stuff anyway, so another post might be in order.
That’s a very tempting thing to try to mess with but also a chunk of stuff I don’t have for testing. Maybe we need another meeting and we can see what we can see?
I’m messing around with https://github.com/n8n-io/n8n and it’s wild what might be possible with this and workflows.
I would love for this to be the guinea pig project, I will message you and we will meet and explore it. Fun, fun, fun!