Welcome to the tXtFL Bowl 1.0! To celebrate our 0.9.1 release and Super Bowl XLI, we're pitting the Chicago Bears vs. the Indianapolis Colts in a seven-day statistical showdown. Each day of the week prior to Super Bowl XLI, we'll play a game of tXtFL Bowl 1.0 and share our results with you.

Our goal is to test our game model...and make some fantasy predictions while we're at it. Feel free to download a copy of tXtFL and start making predictions of your own.

Results.

Methods. Using the all-new graphical Draft for tXtFL, we've downloaded CHI and IND, including all their players. We also downloaded player stats from Draft. Choosing a balance between complete accuracy and a more typical setup, we downloaded only the main statistics for each player type, such as QB passing while ignoring QB receiving. We looked up the teams' depth charts on NFL.com and adjusted the team roster position rankings. In cases where a player was placed on a specific side of the field, we adjusted the player's spec file to reflect this assignment (eg "WR" --> "RWR" for a wide receiver who normally lines up on the right). We made a copy of the players and teams folders so that we could have fresh copies for each game, rather than accumulating statistics from the tXtFL games.

For each game, we recorded the points at the end of each quarter. At the end of the game, we record and highlighted stats of major players before reverting the spec files to original form. A summary includes each game and mean scores from all games. Along the way, we made notes about potential tXtFL bugs, and we intend to continue analyzing the stats to generate more realistic game models.

Analysis. Although the final score put Chicago erroneously—in retrospect—on top, the Indianapolis score breakdown was surprisingly accurate. tXtFL predictions for points gained each quarter differed by no more than 3 points, or 1 field goal, for a total difference of less than 4 points, or under 1 touchdown, from the actual score.

The predictions for Chicago were a little less promising. Scores by quarter differed up to 17 points from the actual score, for a total difference of 58 points. Evidently tXtFL widely and consistently overestimated the ability of the Bears. We looked for possible statistics that might have skewed the results. So far we found a bug in passer rating calculations (now fixed) that reduced some of the advantage that Peyton Manning would have otherwise enjoyed. Subsequent tests with the corrected ratings have brought the scores closer, but have not brought Indianapolis on top, as would otherwise have been expected given their statistics. It is possible that the distribution of players somehow favored Chicago by weighting players in positions more frequently utilized in tXtFL.

A variety of statistics have yet to be incorporated. Kicker distances and field goals have not been translated into player specs. "Second string" players are also ignored, meaning that a team must rely solely on the strength of its "first stringers," skewing against teams that utilize two above average players for a given position rather than a single star. In our tests, we chose not to adjust any coaching points or player skill values to avoid subjective skewing, but the default values cannot take advantage of such intangible qualities as leadership, specific physical strengths, or brains on the field.

An old hangup has been the lengthy clock. The number of passes per game, sometimes exceeding 100 attempts, illustrates that either the clock is slow, or the number of rushing plays is far too low. The fact that no RB exceeded 100 yards rushing in a game indicates that the dearth of rushing plays is at least part of the problem.

Conclusions. tXtFL makes some accurate predictions but has much room for improvement in its stastics input and output. Plans are underway to infuse the Draft utility, now graphical for the first time, with greater statistical processing power and easier to use controls for adjusting subjective player parameters. The game engine is also being updated to evaluate these statistics more accurately, including better utilization of a wider variety of players. The statistical output during game time can also be improved for grammar, clarity, and relevance. A game summary file that tracks all stats and which can be easily added to a spreadsheet for further analysis is also in the works.

Thanks for joining us for tXtFL Bowl 1.0! Keep checking back this spring and summer, and look forward to a season of tXtFL ahead of Bowl 2.0.