Thursday, April 1, 2021

Game 249: Computer Quarterback

I am so unqualified to review this game. I already don't care for sports video games in general, or enjoy anything at all about professional team sports of any kind, and am especially uninterested in football, to the point where I'm not even completely sure what the rules are. I did cover Football Manager (the other kind of football) back in 2019, but the actual play mechanics are really just about balancing your team's skill level and rotating players to manage fatigue. Computer Quarterback, rising superstar Dan Bunten's first and most successful game released through SSI, claims to be a comparatively deep simulation of the sport, based on years of painstaking mathematical analysis. Although SSI's business was in wargames, football, with its field of engagement, its centralized planning and control of coordinated action, its impenetrable (to me anyway) diagrams of X's and O's to depict attack and defense strategies, its endless supply of statistics, and its shockingly high rates of traumatic injury makes the sport well suited to be simulated as one simulates war, the play-calling quarterbacks its front line generals.

As a necessary aside, Bunten would undergo sex reassignment surgery in 1992 and live the rest of her life as Danielle Bunten Berry, but would refer to her past self as "him" on her personal webpage. For the rest of this post, which only concerns events of 1981 and prior, I will too.

Computer Quarterback's roots date back to 1976 when Bunten, a certified industrial engineer, developed a football game simulator in FORTRAN on a minicomputer at the National Science Foundation. In 1978, he designed his first commercial computer game Wheeler Dealers, which included a custom-made peripheral for simultaneous four player bidding action on the Apple II. Sales were abysmal, and there don't seem to be any copies of the game on the Internet or any pictures of its peripheral. For his next product, he would return to his old simulator and construct a playable game around its mathematical modelling of the sport. This sort of detailed modelling and boardgame-like play structure was a perfect fit for SSI's catalog, who insisted on the addition of a computer opponent to play against. It became an early offering, and like several other of SSI's earliest games, was later given a second edition featuring hi-res mode graphics and other enhancements. This second edition, also ported to Atari and Commodore computers, is the only version that seems to still exist.

Scan by Atarimania. I don't know what any of this means!

The crux of the game is plays. During each down, offense and defense both choose a play, and after that the outcome is up to chance. Every combination of plays has a distribution of outcomes - we're told, for instance, that a Z Post against a Key Man has a 28% completion rate with a mean gain of 36 yards, 8% interception possibility, 15% chance of trapping the quarterback, and a 1.5% chance of a fumble. The manual has a chart showing gains and completion rates for the 208 combinations of standard plays, but doesn't break down the distributions further.

The result strikes me as something like a combination of rock-paper-scissors and craps. Think your opponent on the offense might be going for the long shot Flea Flicker? The Prevent defense has a 94% chance of stopping that, and pretty good odds on other passing plays, but it's crummy against screens and runs, though those can still fail spectacularly against any defense.

Not listed here are field goals and punts, two options for offensive plays, and punt returns, an option for defensive play.

I played a match with "B," who knows a bit more about football than me, but is far from being a fan.

Weirdly, Computer Quarterback requires paddles to play, and I'm not entirely sure what the goal here was. You turn the paddle to select a number representing a play. Offense commits to a play by tapping the button three times (each producing a beep, suggesting "hut, hut, hut"), and then you see the plays called and watch the ball move.

By using paddles instead of a keyboard, each player can see what the other one is doing and adjust their tactic. Maybe I'm missing something as someone who doesn't follow football, but I can't see how this does verisimilitude any favors. I had assumed that teams decide their plays in secret. And that once they see how the defense is shaping, that maybe they can make quick tweaks, but certainly not completely change their strategy. But here, all it takes is a nudge of the dial to change your play from "Reverse" to "Z Slant" when you see that your opponent has "Short Yardage" selected.

I'm also not sure what the point is from a pure gameplay perspective. Offense can change their play as a reaction to a defense, but defense can't in the reverse. Once offense commits, it happens, and if defense hasn't also committed, then it will be disorganized and ineffective. To ensure defense has a chance, only defense may commit during the first five seconds of play selection.


The odds do seem to be in offense's favor - mean yardage for almost every combination of plays is greater than the 2.5 that you'd need to average in order to win 10 yards every four downs, but either "B" is much better at this game than me, or I got very unlucky. Possibly both. He would consistently meet or beat the odds, while I would usually come up short, even when I picked the optimal (according to the chart) attack for the defense I was going up against.

Mean yardage of this play is 8, but I only got 1, and I needed 3.

Penalties against my team were also more frequent and more severe. The other team always has a choice to refuse a penalty, but I'm not sure why you would do that.

Manatees score a touchdown.
Your halftime show is math.

Flea Flicker is high risk.

But also high reward.

My only touchdown.

Ahead 12 points, "B" runs out the clock.

"B" said he had more fun with this than he expected to, but also felt like his win was more from luck than skill. He did play with more of an anticipatory style, while I played in a more reactive way, which in theory should have given me an advantage on offense, but evidently the theory is either wrong or it's subject to luck that just wasn't on my side.

Computer Quarterback has some options that we didn't and won't delve into.

  • Team drafting, in which you can assemble a team of players with different stats on an agreed-upon budget (or a dream team with unlimited budget). Stats will change the effectiveness of plays, and can optionally be done in secret to keep your opponent from knowing your strengths and weaknesses. SSI would later sell NFL roster disks for 1984 through 1986.
  • An option to randomize or manually choose the numbers corresponding to plays. The intent is to make your plays secret, but the idea seems flawed - your opponent might not know what it means when you have "1" selected as your offensive play the first time, but after that it would become revealed and no longer be secret. If the only goal was to make plays secret, why not just allow the keyboard, and keep the typed numbers hidden?
  • A "professional" mode, offering "36 offenses, 24 defenses, 3 alignments, & double teaming of receivers." I don't even want to try to guess how that all works - the manual offers no result chart equivalent for this mode, so I guess you just have to know football to have any sort of inkling when you'd want to use an "Action X Square Out" as opposed to a "Y Cross Medium."
  • An obligatory singleplayer mode against the "Robots." This uses a data disk that gets saved to between games, so that, according to the game, they can "learn from you." Professional play is not allowed in singleplayer.

GAB rating: N/A. Like I said, I don't feel qualified to cover this game. It wasn't meant for me, and even after playing my grasp on how things work is loose at best. So I won't rate it.

Bunten's notes do say that Computer Quarterback was conceived as a simulator first, and that the "game" aspect was more of an ad-hoc addition, albeit one that took over a year to implement. As simplistic as the interactive elements might be, they're oddly anticipatory of M.U.L.E., in that a simple input device allows simultaneous multiplayer action, with incredibly complicated math going on behind the scenes to dictate the results. In any event, SSI's playerbase seemed to like it.


  1. Speaking as someone who does like football a lot, a few comments:

    >Offense can change their play as a reaction to a defense, but defense can't in the reverse. Once offense commits, it happens, and if defense hasn't also committed, then it will be disorganized and ineffective.

    This does reflect the way the real game works. The offense starts the play with the snap, so they are able to "audible" (change the play) in response to how they see the defense line up, and it's harder for the defense to respond in kind.

    >The other team always has a choice to refuse a penalty, but I'm not sure why you would do that.

    The basic component of football is that you have 4 chances to go 10 yards, and if you do, you get a new set of 4 chances (downs). If there is a penalty on the offense, the offense has to move back on the field, but they (usually) get to repeat the down. So the defense might decide to decline the penalty so that the offense doesn't get to repeat the chance.

    For the offense, if they accept a penalty against the defense they also usually get to retry the down -- but if the play went well, they may want to accept the result of the play instead of the penalty.

  2. You decline a penalty when the result of the play is better. For example, if the defense gets an interception, it would rather have that result and decline a penalty for offensive holding.

    The flea flicker is a trick play. Trick plays usually don't work, but when they do work, the results can be spectacular as the defense wasn't expecting it.

    1. Suppose defense is a 3-4 Key Zone. The chart tells me Flea Flicker has a 39% completion rate against that, and math tells me that four attempts would have an 86% chance of at least one success (including a non-zero chance of multiple successes). On paper that seems a lot better than trying to make it 10 yards at a time, with so many chances for things to go badly (penalties, fumbles, consecutive poor gains).

      In fact, passes seem like they would get better overall yardage than runs against all defenses, if we calculate expected yardage by multiplying average gains by completion rate. Deep Zone supposed to stop passes, but the best run has an average gain of 6, while HB Option has average of 18 and a more than 33% completion rate, giving it a higher expected yardage.

      But sometimes intuition beats cold number analysis, I suppose. Plus the chart doesn't tell you everything - no interception rates, for instance, so the math could certainly be misleading. In any event, it didn't secure my win. *cue John Nash's hissy fit*

    2. I can tell you have a surface-level knowledge of the game. There are three things that can happen when you throw the ball, and two of them are bad. If you're calling trick plays more than once or perhaps twice a game, you're doing it wrong. Joe Theismann's career famously ended when the defense was not fooled by the flea flicker and Lawrence Taylor crushed his leg tackling him.

      On an incomplete pass, the clock stops. On a run, the clock keeps ticking. A very viable strategy is to hog the ball for as much time as possible, grinding out three yards at a time, thus giving the opponent less time on offense. This also tires out the defense by keeping them on the field, but I don't know if this game models fatigue.

      An interesting experiment would be your robot analysis method against someone who actually knows how to play football. Three games would be enough to settle the question. A pity it's not worth spending the time to find out.

    3. Of course there's a difference between real football and this video game -- it may very well be effective to call flea flickers every play in Computer Quarterback. Other games like Madden are also notorious for having unrealistic success rates on plays.

      (I had a computer football game for DOS BASIC as a kid where you could make a 97 yard field goal)

    4. But this football game is supposed to be a *simulation*, not a rock 'em, sock 'em game like Madden football.

    5. In the manual's designer notes, Bunten boasts of the mathematical rigor that went into the ball game model. I guess that to accurately model the effectiveness of trick plays, you'd need to implement some sort of rapidly diminishing returns on the odds to show that the flea flicker isn't fooling anyone the third time you use it. And it's not unrealistic to imagine that this could be done even in 48KB BASIC. But then there's a problem - an accurate model needs lots of data, and if nobody in real life uses a trick play more than twice, then there's no data showing what happens when you do. You'd have to extrapolate from the data available.

      The anecdote of the 97 yard field goal immediately makes me think of Tecmo Bowl, of which the only thing I know about is that Bo Jackson can easily run 99 yards with a cloud of defenders chasing after him and eating astroturf, then turn around and run all the back to the scrimmage line before turning around again and scoring a touchdown.

    6. It would surprise me if the "mathematical rigor" refers to the dynamics of the simulation at runtime rather than the construction of the static probability tables that the simulation uses.

  3. Wow, didn’t know such an early gaming pioneer was transgender! You made my day!


Most popular posts