Data Does Not Lie: Week Four 2026

New Series: Data Does Not Lie

I’m old and tired now. Grad school and work make it almost impossible to write power rankings that shit on your team every week. Instead, I’m just going to let the data tell the full story here periodically. In case you missed it, I stole Callen’s idea to build a Sleeper lakehouse using Databricks Free Edition. All data comes from there, and you can easily run all code snippets below in a Databricks notebook. I’m happy to grant you access. Just share your email address. It’s time to let the hate flow.

Luck and Expected Wins

When I previously analyzed luck in the league’s history, I found that $w_o$, the number of teams that your score would beat on any given week, is arguably the best performance metric. This has been built into the lakehouse as expected_wins and falls into a range between 0 and 1 using min-max scaling. In short, if you get the highest score, your expected wins are 1. If you have the worst, then it’s 0. The same idea, but we can also quantify luck relative to actual wins. I’ll start by ranking everyone by expected wins:

season-summary.sql

with totals as (
    select manager_display_name as Manager,
           sum(points) as PF,
           sum(opponent_points) as PA,
           sum(expected_wins) as ExpectedWins,
           sum(case result when 'win' then 1 when 'tie' then 0.5 else 0 end) as ActualWins
    from sleeper.gold.weekly_scores
    where season = 2026 and week < 5
    group by manager_display_name
)
select Manager, round(PF, 2) as PF, round(PA, 2) as PA, round(ExpectedWins, 2) as ExpectedWins, ActualWins,
       round(ActualWins - ExpectedWins, 2) as Luck
from totals
order by ExpectedWins desc

Let’s see how that looks:

Manager	PF	PA	Expected Wins	Actual Wins	Schedule Luck
coolguytrav	553.00	448.17	3.82	3.0	-0.82
exoentropy	502.38	402.21	3.00	4.0	1.00
lrich91	469.87	409.11	2.55	3.0	0.45
BigOleDoinks	463.86	445.02	2.45	2.0	-0.45
novaprice	455.56	376.04	2.36	4.0	1.64
markhambone	436.67	491.19	2.18	2.0	-0.18
mattkman	415.84	463.71	1.64	1.0	-0.64
callentrail	408.05	457.93	1.64	1.0	-0.64
almostwinning	371.58	367.44	1.18	2.0	0.82
cramseyio	407.33	436.97	1.18	1.0	-0.18
jyarrow	356.07	450.18	1.09	1.0	-0.09
acarlough	362.43	454.67	0.91	0.0	-0.91

Tarv’s team is a country mile ahead of everyone in PF. His not being 4-0 is squarely due to bad luck. His one loss is the classic scenario of scoring second PF for the week while playing the top scorer. A sub-one-point loss on top of that is just brutal. This team is still the best, no question. On the flip side, Caleb is by far the largest beneficiary of luck. Standing in fifth in PF and 4-0 is only possible because of an extra 1.64 wins from pure luck. The most egregious example is last week, when his team finished third-to-last in points but happened to play one of the two teams with even fewer. This stat won’t capture it, but Caleb’s luck continues by playing me when my QB, RB1, and RB3 are all on bye. Yours truly is the other beneficiary of luck. While I’m second in PF by a comfortable margin, my luck came from playing Carlos during my only down week.

Keeper Value

It’s obviously too early to analyze everyone’s draft decisions and strategies (e.g., a few high-end players, several mid-tier players, etc.). That said, we can review points per dollar on an ongoing basis:

keeper-value.sql

select player_name as Player, position as Position, nfl_team as Team,
  manager_display_name as Manager, amount as Cost, points_for_drafting_team as PF,
  round(points_per_dollar, 2) as PointsPerDollar
from sleeper.gold.draft_spend
where season = 2026 and is_keeper = true
order by PointsPerDollar desc;

Here’s the output:

Player	Position	Team	Manager	Cost	PF	Points per Dollar
Javonte Williams	RB	DAL	coolguytrav	14	80.4	5.74
Chris Olave	WR	NO	mattkman	20	79.1	3.95
Kyren Williams	RB	LAR	novaprice	20	76.7	3.84
Kenneth Walker	RB	KC	exoentropy	29	101.1	3.49
Emeka Egbuka	WR	TB	exoentropy	14	40.1	2.86
Cam Skattebo	RB	NYG	callentrail	14	36.2	2.59
Amon-Ra St. Brown	WR	DET	markhambone	29	72.3	2.49
Brock Bowers	TE	LV	almostwinning	17	42.2	2.48
Chase Brown	RB	CIN	coolguytrav	17	40.0	2.35
Tyler Warren	TE	IND	mattkman	14	29.5	2.11
Ladd McConkey	WR	LAC	cramseyio	17	33.3	1.96
Drake Maye	QB	NE	callentrail	14	25.7	1.84
Rashee Rice	WR	KC	cramseyio	20	30.0	1.50

Callen thought Javonte Williams was obviously the best keeper during the draft. This has held so far in terms of ROI. Kenneth Walker is the best keeper from a total PF perspective. Caleb continues to lean on Kyren Williams. Keepers clearly bolster the top teams on the leaderboard. Callen and Cramsey have not-so-great ROI. Drake Maye was already dropped, and Cam Skattebo continues to deal with injury issues while averaging 0.75 fumbles per game. Ladd McConkey and Rashee Rice are currently WR 39 and 43, respectively. Not terrible, but not worth the price tag.

Flyers

We’ll look at the high-end players later, but I was curious whose one-dollar picks went the farthest. Yes, I know it’s an arbitrary cutoff. I also excluded QBs (not because I don’t want Patrick Mahomes on any of my lists, I swear) and defenses. We can tweak the query above:

flyers.sql

select player_name as Player, position as Position, nfl_team as Team,
  manager_display_name as Manager, amount as Cost, points_for_drafting_team as PF,
  round(points_per_dollar, 2) as PointsPerDollar
from sleeper.gold.draft_spend
where season = 2026 and amount = 1 and position not in ('QB', 'DEF') and points_for_drafting_team > 0
order by PointsPerDollar desc;

Look at that! We’ve got a Longhorn legend at the top:

Player	Position	Team	Manager	Cost	PF	Points per Dollar
Matthew Golden	WR	GB	novaprice	1	51.0	51.00
Deebo Samuel	WR	SF	lrich91	1	45.9	45.90
Isaiah Likely	TE	NYG	novaprice	1	41.0	41.00
Jakobi Meyers	WR	JAX	almostwinning	1	36.7	36.70
Wan’Dale Robinson	WR	TEN	jyarrow	1	32.4	32.40
Josh Downs	WR	IND	almostwinning	1	32.2	32.20
Jordan Addison	WR	MIN	acarlough	1	28.2	28.20
Alvin Kamara	RB	NO	lrich91	1	25.1	25.10
Keaton Mitchell	RB	LAC	markhambone	1	23.8	23.80
Juwan Johnson	TE	NO	BigOleDoinks	1	22.3	22.30
Rachaad White	RB	WAS	jyarrow	1	22.1	22.10
Romeo Doubs	WR	NE	novaprice	1	17.5	17.50
Rashid Shaheed	WR	SEA	lrich91	1	17.0	17.00
Jonah Coleman	RB	DEN	novaprice	1	11.8	11.80
Tyler Allgeier	RB	ARI	acarlough	1	10.8	10.80

Caleb clearly shines here. Trondheim also has some good value picks. Mark and John probably planned on being higher on this list, given the top dollar they spent on their core.

Top-Dollar Talent

Next, I wanted to know more about the premium picks. Before diving in, I first defined what ‘premium’ means using some simple Spark code.

statistics.py

df = spark.table('sleeper.gold.draft_spend').where('season = 2026')
display(df.select('amount', 'points_for_drafting_team', 'points_per_dollar').summary())

Statisticians rejoice.

Summary	Amount	Points for Drafting Team	Points per Dollar
Count	156	144	144
Mean	15.37	37.10	9.04
Standard Deviation	19.81	26.12	14.38
Minimum	1	-0.60	-0.60
25th Percentile	1	17.50	1.37
Median	6	34.30	3.23
75th Percentile	21	47.80	8.08
Maximum	88	116.36	91.78

Great! Let’s use that 75th percentile to define premium:

high-end.sql

select player_name as Player, position as Position, nfl_team as Team,
  manager_display_name as Manager, amount as Cost, points_for_drafting_team as PF,
  round(points_per_dollar, 2) as PointsPerDollar
from sleeper.gold.draft_spend
where season = 2026 and amount >= 21
order by PointsPerDollar desc
limit 15;

I was on the fence about whether to include QBs here. I ultimately kept it because Josh Allen is the top player overall by fantasy points. For the record, I think Jalen Hurts is only on here because of intrinsic position value.

Player	Position	Team	Manager	Cost	PF	Points per Dollar
Josh Allen	QB	BUF	BigOleDoinks	32	116.36	3.64
Kenneth Walker	RB	KC	exoentropy	29	101.10	3.49
Jalen Hurts	QB	PHI	lrich91	23	72.82	3.17
Tee Higgins	WR	CIN	coolguytrav	21	65.10	3.10
Amon-Ra St. Brown	WR	DET	markhambone	29	72.30	2.49
Tetairoa McMillan	WR	CAR	almostwinning	27	62.50	2.31
Zay Flowers	WR	BAL	coolguytrav	28	61.20	2.19
Bucky Irving	RB	TB	lrich91	29	63.00	2.17
Garrett Wilson	WR	NYJ	novaprice	26	51.00	1.96
CeeDee Lamb	WR	DAL	lrich91	58	100.10	1.73
Jaxon Smith-Njigba	WR	SEA	coolguytrav	64	103.76	1.62
Derrick Henry	RB	BAL	exoentropy	54	85.80	1.59
Quinshon Judkins	RB	CLE	cramseyio	26	38.30	1.47
DeVonta Smith	WR	PHI	acarlough	27	38.50	1.43
David Montgomery	RB	HOU	mattkman	28	38.40	1.37

There’s an argument to be made (at least so far) that Kenneth Walker is the best pick in the draft. The highest flex player by points per dollar in the top quartile. But what about the very top by spend? Let’s check that out too:

top.sql

select player_name as Player, position as Position, nfl_team as Team,
  manager_display_name as Manager, amount as Cost, points_for_drafting_team as PF,
  round(points_per_dollar, 2) as PointsPerDollar
from sleeper.gold.draft_spend
where season = 2026 and amount >= 21
order by amount desc
limit 15;

The results are a bit surprising:

Player	Position	Team	Manager	Cost	PF	Points per Dollar
Jahmyr Gibbs	RB	DET	BigOleDoinks	88	102.00	1.16
Bijan Robinson	RB	ATL	markhambone	87	91.40	1.05
Jonathan Taylor	RB	IND	cramseyio	71	79.20	1.12
Ja’Marr Chase	WR	CIN	jyarrow	70	47.20	0.67
James Cook	RB	BUF	novaprice	66	61.50	0.93
Puka Nacua	WR	LAR	mattkman	65	33.10	0.51
Jaxon Smith-Njigba	WR	SEA	coolguytrav	64	103.76	1.62
Christian McCaffrey	RB	SF	BigOleDoinks	61	65.00	1.07
Saquon Barkley	RB	PHI	jyarrow	61	19.00	0.31
De’Von Achane	RB	MIA	almostwinning	59	19.60	0.33
CeeDee Lamb	WR	DAL	lrich91	58	100.10	1.73
Justin Jefferson	WR	MIN	jyarrow	58	37.90	0.65
Derrick Henry	RB	BAL	exoentropy	54	85.80	1.59
Omarion Hampton	RB	LAC	novaprice	53	41.70	0.79
Ashton Jeanty	RB	LV	lrich91	51	59.90	1.17

This is far from a perfect metric, but I’m shocked that points per dollar hovers around 1 for the top picks. There are obviously benefits not captured here (e.g., positional scarcity, week-to-week consistency, etc.), but this illustrates the premium paid in the draft for top-end dudes.

Closing Thoughts

I’m starting to like auction draft more and more as the season goes on. Being able to track ROI and efficiency against a dollar amount you specified (rather than an arbitrary position order) unlocks a lot of fascinating analytical potential. I feel like I’ve only scratched the surface of the statistical shit-talking this data unlocks. It’s beautiful.

Tarv is the team to beat this year and resides in his own tier. The only teams that I’m writing off are Carlos and Trond. Everyone else is in the hunt.

Until next time, tryhards.