Eight weeks into the first season of the Inaugural League, Promethium has a real data set. Sixteen managers, 64 head-to-head matchups, 1,152 draft picks, and every one of them made from the roster of a single real NFL game. That last part is what makes this different from any fantasy football data I have ever looked at. In a normal league your team is spread across 14 games on a Sunday. Here, both teams in a matchup come out of the same game, so every number below is really a question about that game: how many points it produced, how they were split, and who ended up holding them.
This post is a deep dive into what the first half of the season says about the NFL when you look at it one game at a time. This is about the games, the positions, the draft and the bot.
A quick note on how the points are counted: everything here is scored under the Inaugural League's rules. A point per reception, 6 points for a passing touchdown, and no penalty for an interception or a lost fumble. A roster is nine players: a quarterback, two running backs, two wide receivers, a tight end, a flex, a kicker and a team defense, so a draft is 18 picks.
The shape of a single-game matchup
The average team scored 91.5 points a week, and the average matchup produced 183.1 combined. The typical result is not close: the median margin was 21.9 points and the average was 26.3. But the spread is wide, and the tails are where the fun is.
- 12 matchups (19%) were decided by fewer than 5 points.
- 20 matchups (31%) were decided by single digits.
- 16 landed between 10 and 25 points.
- 19 were 25 to 50 point games.
- 9 were blowouts of 50 or more.
The closest of the season was decided by half a point, out of the Saints at Ravens game in Week 2, and it came down to a Lamar Jackson interception on the final play. The biggest margin was 90.2 points, out of Cowboys at Texans in Week 4.
The real game sets the ceiling
The first thing I wanted to know was whether a high-scoring NFL game makes for a high-scoring fantasy matchup. It does, and it is not subtle.
Across all 64 matchups, the real game's combined score and the fantasy matchup's combined score have a correlation of 0.79. For a sport this noisy, that is enormous. On average, every real point scored in a game turned into 4.1 fantasy points split between the two rosters.
The extremes line up exactly the way you would expect:
- Titans at Bengals, Week 8 finished 43-36 in overtime. It produced 282 fantasy points, the most of any matchup this season.
- Lions at Bills, Week 2 finished 41-31 and produced 253.5.
- Bills at Rams, Week 5 finished 37-34 in overtime and produced 247.2.
At the other end:
- Falcons at Buccaneers, Week 8 finished 16-0. One offense never scored, and the matchup produced 97.5 fantasy points combined, the fewest of the season.
- Vikings at Bears, Week 2 finished 9-3 and produced 126.7.
If you know the over/under, you know roughly how many points are on the table before anyone picks.
Overtime deserves its own mention. Seven of the 64 drafted games went to overtime, and three of them (Titans-Bengals, Bills-Rams, Colts-Chiefs) are among the higher-scoring matchups of the season. An extra period is free fantasy points for everyone, and it tends to land in games that were already shootouts.
But the real scoreboard does not pick the winner
Here is the result that surprised me most. The real game's margin has almost nothing to do with the fantasy margin. The correlation is -0.14: essentially zero, and if anything slightly backwards.
It gets stranger when you split the games up:
- In real one-score games (decided by 7 or fewer, 36 of them), the fantasy matchups were decided by an average of 30.1 points.
- In real blowouts (14 or more, 21 of them), the fantasy matchups were decided by an average of 21.7 points.
The nail-biters on the field produced the lopsided fantasy results, and the blowouts produced the closer ones.
My best explanation: in a close real game, both offenses keep their starters in and keep throwing, so the points concentrate in a few stars, and whoever drafted those stars runs away with it. In a real blowout, the losing team abandons the run and the winning team pulls its starters, and the points get spread across more players and more garbage time. Spread-out points make for closer drafts.
The winning NFL team does produce more fantasy points: 101.7 per matchup from players on the real winner, against 80.8 from the real loser. The average roster drew 56% of its points from the winning side. But the manager who held more of the winning team only won 54% of the time. Picking the right side of the real game is a small edge, not a strategy.
Which NFL teams make the best matchups
Every NFL team has now been drawn at least once. The Falcons, Ravens and Bills have each been drafted from 7 times, and the Eagles just once. Below, the teams with at least three games drawn, ranked by what their games produced. These are small samples, so treat them as trends and not laws.
The best fantasy environments (average combined points in matchups drafted from their games):
- Cincinnati: 228.4 (real games averaging 62.3 total points)
- Detroit: 218.1 (53.4)
- Buffalo: 213.0 (59.1)
- Las Vegas: 201.0 (57.5)
- Los Angeles Rams: 200.0 (54.3)
- Dallas: 197.7 (52.4)
Cincinnati's own players alone have averaged 136.2 fantasy points a game when the Bengals are drawn. That is more than an entire average team. Detroit's own players average 118.3.
The leanest environments:
- Tampa Bay: 153.0 (real games averaging 42.2)
- Minnesota: 153.5 (28.3, the lowest real totals of any team)
- Miami: 155.9
- New England: 156.9
- Seattle: 163.1
The closest matchups, by average fantasy margin:
- New Orleans: 2.9 points. Three Saints games drafted, decided by 0.5, 1.6 and 6.5 points. If you want a nail-biter, pray for the Saints.
- New York Jets: 11.1
- Miami: 12.2
- Jacksonville: 14.9
The most lopsided:
- Cleveland: 49.9
- Carolina: 40.7
- Seattle: 37.9
- Indianapolis: 37.5
Notice that Miami is on both the lean list and the close list. When a game does not produce much, there is not much to separate two rosters. The big environments cut the other way: more points on the board means more room for one roster to pull away.
Where the points come from
By position, across all 1,152 roster slots:
- QB: averages 20.5 points, 22% of all points scored. Only 5% scored zero or less.
- WR: averages 10.3, but there are more of them, so receivers produced 29% of all points, the largest share of any position.
- RB: averages 9.8, 23% of points. 13% scored zero or less.
- TE: averages 7.7, 11% of points.
- K: averages 8.0, 9% of points. Only 3% of kickers were shut out, which makes the kicker the most dependable slot on the board.
- DEF: averages 5.7, 6% of points, with a range from 24 down to -3.
The more interesting question is which positions decide matchups. For every game with a winner, I compared the winner's and loser's points at each position:
- WR: the winner averaged 14.2 more points, and had more in 78% of games
- QB: +4.5, more in 64% of games
- RB: +5.6, more in 61%
- K: +0.4, more in 53%
- DEF: +0.7, more in 52%
- TE: +0.8, more in 50%
Games are won at wide receiver. The manager with more receiver points won more than three quarters of the time. Tight end, kicker and defense are coin flips: the winner was about as likely to lose those slots as win them.
The quarterback finding is interesting because of what is missing. In a single-game draft, there are only two starting quarterbacks, and each roster gets one. Having the better quarterback won 64% of the time. That is a real edge, but not the edge you would guess, because the receivers he throws to are on the board too, and both managers are fighting over them.
Everyone stacks, whether they mean to or not
In season-long fantasy, "stacking" a quarterback with his own receiver is a deliberate, high-variance strategy. In a single-game draft it is not a strategy at all. It is unavoidable.
Every one of the 128 rosters drafted this season had its quarterback paired with at least one of his own pass catchers. With only two teams' worth of receivers available and two to three receiver slots per roster, you cannot avoid it. That is part of why receivers decide games: when your quarterback throws a touchdown, you very often score on both ends of it.
The draft: what picks are worth
Average points by round, two picks a round:
- Round 1: 18.7
- Round 2: 13.9
- Round 3: 11.4
- Round 4: 9.8
- Round 5: 8.1
- Rounds 6-8: about 8 (this is where kickers and defenses go, and they hold their value)
- Round 9: 5.9
The first overall pick averaged 20.0 points, the second 17.4. The first pick out-scored the second in 38 of 64 drafts (59%). But the first overall pick was the single highest scorer in the matchup only 12 times out of 64. The best player in a game is usually not the one everyone agrees on.
What went first overall: a running back 27 times, a quarterback 18 times, a wide receiver 18 times, and, once, a team defense. I respect it.
Does picking first help? The manager with the first pick won 34 of 64 matchups, which is 53%. That is a slight edge, and well inside what you would expect from chance over this many games.
On average, positions came off the board in this order: running backs (pick 7.5), wide receivers (7.9), quarterbacks (8.7), tight ends (10.3), kickers (13.2) and defenses (13.9).
The late-round quarterback is the best value in the game. Because every roster needs exactly one QB, the second quarterback in a game often lasts deep into the draft, and he routinely out-scores the players taken ahead of him. The four best performances from the last four picks of a draft were all quarterbacks:
- Drake Maye: 36.2 points as the 18th and final pick
- Kirk Cousins: 34.2 at pick 16
- Jameis Winston: 33.7 at pick 16
- Bo Nix: 30.1 at pick 18
If you are drafting second at quarterback, you may be getting the better deal.
The best single performance of the season was Joe Burrow's 55.5 points in that Titans-Bengals shootout, taken fourth overall. After him: Josh Allen with 46.8 (first overall), Tetairoa McMillan with 45.2 (ninth pick), Jaxon Smith-Njigba with 42.5 and CeeDee Lamb with 41.3.
The bot
When a manager is not in the draft room, the draft bot picks for them, ranking players by their projected points over a replacement-level player at the same position. It has been busy.
The bot made 421 of 1,152 picks: 37%. That share has climbed all season:
- Week 1: 23%
- Week 2: 14%
- Week 3: 32%
- Week 4: 42%
- Week 5: 38%
- Week 6: 31%
- Weeks 7 and 8: 56%
More than half of all picks in the last two weeks were made by the bot. Some of that is a busy stretch of the calendar, and some of it, I suspect, is managers deciding the bot is good enough.
Is it? Per pick, the bot averaged 10.3 points. Humans averaged 10.1. By position:
- RB: bot 10.2, humans 9.6
- K: bot 8.2, humans 7.8
- DEF: bot 6.0, humans 5.5
- QB: bot 20.3, humans 20.6
- WR: bot 10.1, humans 10.4
- TE: bot 7.0, humans 8.1
The bot is slightly better at the positions where projections are most reliable (running back volume, kickers, defenses) and worse at tight end, where knowing who actually gets the targets this week matters most.
When whole rosters are compared, humans hold a small edge. Rosters drafted entirely by the bot won 49% and averaged 91.6 points. Rosters drafted entirely by humans won 53% and averaged 93.2. In the 13 matchups that pitted a fully bot-drafted roster against a fully human-drafted one, the bot won 6 and the humans won 7. That is as close to a tie as you can get.
The strangest category is bot against bot: matchups where neither manager showed up and the bot drafted both rosters. There have been 16 of them, a quarter of all matchups. You would expect a machine drafting against itself to produce dead-even results. It does not. The bot-versus-bot games include a 1.6-point finish (Falcons at Saints), but also two of the three biggest blowouts of the season, by 81.5 (Lions at Panthers) and 79.8 (Browns at Titans). Both rosters were built from the same projections, so the gap is whatever the projections could not see: the game itself.
The bot made the first overall pick 24 times: a quarterback 11 times, a wide receiver 7, a running back 6. It also made the season's best pick, Joe Burrow at fourth overall for 55.5 points, and one of its worst: Tua Tagovailoa at third overall for 0 points.
Odds and ends
- Injury swaps were used just 5 times in 64 matchups. Most of the time, the player you draft plays.
- Best kicker game: Evan McPherson, 19 points. Worst: two kickers scored 0.
- Best defense: Tampa Bay, 24 points in a 43-13 win at Carolina. Worst defense: Carolina's, at -3, in that same game. One game, both ends of the scale.
- The Falcons, Ravens and Bills have been drawn more than anyone (7 times each). The Eagles have been drawn once.
What I am taking from this
A few things I did not know eight weeks ago, and what I plan to do about them.
The real game's total matters more than anything else on the board. A 0.79 correlation between the real score and the fantasy score means the over/under is effectively a projection for the whole matchup. I want to surface that in the draft room, so you can see before the draft whether you are walking into a Bengals shootout or a 9-3 Vikings game.
Wide receivers decide games. The projections, and the bot that relies on them, should reflect that.
The bot is good enough that a lot of you are letting it draft for you. It is almost dead even with humans pick for pick. That is a compliment to the bot and a challenge to everyone else: the edge is there, at tight end and at wide receiver, if you show up for it.
And quarterback is not what it looks like. The second quarterback in a game is the best value in the draft, because every roster needs one and there are only two to go around.
Everything here comes from the Inaugural League, and you can dig into the season's numbers yourself on Player Stats and Promethium by the Numbers. The second half of the season starts now, and with college football (FBS and FCS) now on the site as well, the next one of these is going to have a lot more games in it.
Thanks for reading, and seize the draft.