Anatomy of a champion

Twenty-four champions, each measured against the other thirteen teams in its own league that year. Same season, same scoring, same rulebook.

You do not draft a champion. You build one. Of the eight players who start a championship game, 3.7 of them were not on the roster on draft day — more than were drafted (2.6) and more than were kept (1.6). The median champion's title-game eight were not all on the roster together until week 16. These teams were not finished until the week they won.

And that is where the skill is. Three things predict a season about equally well — your picks beating what those picks are worth, the players you add after the draft, and starting the right eight. Only two of them repeat from one year to the next within the same coach, and both are in-season. Draft night's contribution is real, large, and a coin flip.

The whole edge is at running back — and it is a draft hit, not a plan

Points above the field median at each position, split into starting MORE of them and getting MORE from the ones you started.

Points above the field median at each position, over the 23 champions with player rows. The bar is the total edge; the label splits it into volume (starting more) and quality (scoring more per start).
018355370RB+72 pts volume +12, quality +60QB+13 pts volume −0, quality +13WR−8 pts volume −12, quality +4K+4 pts volume −0, quality +4DEF−2 pts volume −0, quality −2TE+0 pts volume −1, quality +1

Champions beat the field median by 72 points at running back and by essentially nothing anywhere else — and 60 of those 72 points are QUALITY, not usage. They did not start more backs than anyone. The backs they started scored more (z=+0.88, p=0.006).

And the backs came off the draft board. Split by where the player came from, a champion's DRAFTED running backs scored 12.48 per start against the field's 9.58. Its ADDED running backs scored 8.38 against the field's 8.51 — nothing. Champions were no better than anyone at picking running backs off waivers. They were much better at the ones they drafted, having spent no more capital there than their field (capital share z=-0.09). They did not target running back. They hit on it.

Which is why nobody does it twice. Hitting on draft picks is the one thing on this page that predicts a season strongly and does not repeat at all. The single position that decides titles here is decided by the one skill that turns out not to be one.

Six positions were tested, so running back's p=0.006 does not survive correction across them. It is much the largest effect on the page, it agrees with the draft-execution finding and with the origin split, and it is still a lean rather than a proof.

Everything before week 1 is a coin flip. Everything after it is a skill.

A thing a coach does carries into next season. A thing that happened to him does not. The line falls in exactly one place.

whatpredicts scoring champions vs fieldrepeats next year
who you add after the draftr=+0.36+0.51 sd, p=0.02r=+0.15, p=0.03SKILL
who you startr=+0.38+0.27 sd, p=0.07r=+0.23, p=0.00SKILL
your picks beating their slotr=+0.34+0.56 sd, p=0.04r=−0.05, p=0.37UNMEASURABLE
your keepers beating their costr=+0.25+0.39 sd, p=0.09r=−0.04, p=0.75luck
staying healthyr=−0.15-0.71 sd, p=0.00r=+0.07, p=0.23luck

The two highlighted rows are the ones a coach carries with him. Persistence is the same coach's within-season z in consecutive seasons — 266 to 285 pairs — with a gap year breaking the pair.

One row says UNMEASURABLE, and the reason this page used to give for it was wrong. It reported that odd and even halves of the same draft anti-correlate (r=-0.29, 330 team-seasons) and concluded that ten to fourteen picks cannot measure a drafter. The number is real. The diagnosis was not, and it mattered: it let "we cannot measure this" stand in for a question that is in fact measurable. The cause is positional composition, not sample size. A team's picks are not exchangeable - every roster drafts about one quarterback, and with six-point passing touchdowns a quarterback is an outlier in RAW points, so whichever half of a split holds him runs high and the other runs low by construction. A random split reproduces the same negative number, which is the tell: if the scheme were at fault, randomising it would have repaired it. Price each pick against players at its OWN position instead, in position-season SD units, and the anti-correlation disappears - split-half goes from -0.816 to -0.13 on the SAME split scheme. Zero, not negative. (Quoting the random split's -0.028 against the old ALTERNATE figure would compare two different cuts and overstate the repair; the like-for-like pair is the one above.) The two in-season measures split cleanly in half (r=+0.512 and +0.556, full-season reliability 0.677 and 0.715), so their persistence means what it appears to — corrected for that reliability their true year-over-year figures are about 0.28 and 0.25 rather than 0.19 and 0.17.

So: is drafting a skill here? Measurable, consequential, and not demonstrably repeatable. Those are three separate findings and collapsing them is what went wrong the first time. It is measurable - every live pick is priced against its own position, exactly. It matters more than anything else on this page: execution in rounds 1-4 correlates r=+0.41 with where a team finishes, above lineup efficiency and above in-season acquisition. And the band where sleeper-hunting is supposed to live is the weakest one, not the strongest:

picksvs where the team finishedsame coach, next season
rounds 1-4r=+0.41, p=0.000, of 162 team-seasonsr=+0.01, p=0.89 (86 pairs)
rounds 5-9r=+0.18, p=0.008, of 330 team-seasonsr=−0.01, p=0.86 (286 pairs)
rounds 10+r=+0.09, p=0.122, of 324 team-seasonsr=−0.01, p=0.81 (277 pairs)
every roundr=+0.30, p=0.000, of 330 team-seasonsr=−0.04, p=0.44 (286 pairs)

Nothing repeats, in any band. And the strongest test is the career one, because pooling a coach's whole record averages out the single-season noise: across the 22 coaches with five or more drafts, the spread in career execution is what reshuffling the same team-seasons produces 20% of the time. The ranking's own shape says the same thing - the extremes are short careers (Janeen Kopale on 5 seasons at the top), while everyone with twenty-plus converges on zero.

THREE caveats this owes, all of which cut against the number above. The r=+0.41 is partly mechanical - points scored by a team's drafted players are most of that team's points, so the two can hardly fail to correlate. And most of it is not about the SLOT. Zeroing the slope in the per-position fit - so a pick is scored purely against its position's mean, with no information about where it was taken - still returns r=+0.37 of that r=+0.41. So the honest reading of this row is "your drafted players scored well for their position", which is a weaker and different claim than "you beat your slot"; the slot term contributes about a tenth. And the persistence nulls carry a power bound: 286 coach-season pairs resolve r≈0.16, not r≈0.08. The defensible claim is that any repeatable drafting edge in this league is too small for 24 seasons to see - not that it is zero.

And one caveat this page owes on its own number: draft execution's r=+0.34 with team scoring is substantially mechanical. Points scored by a team's drafted players are most of that team's points, so the two can hardly fail to correlate. Read it as a description, not as evidence that drafting well causes winning.

The split is not subtle and it is not where anyone expects it. The two decisions taken before a down is played — which picks to spend and which players to keep — predict a season as strongly as anything else here, and neither repeats. The two taken during it repeat. Preparation matters and does not compound; management matters and does.

Keeping is a good deal, and everyone knows it. Across 151 team-seasons the median keeper beat the pick it cost by 0.27 of an average drafted player, and 65% of keeper decisions came out ahead of the round they consumed. Champions were no better at it than their field (). Asking "how many keepers did champions have" could never have found this: the count is capped at four and most teams sit at the cap, so there is no variance in it to spend.

Notice what is missing from that table: the draft board. Twenty-two separate measures of how a team drafted — how many of each position, what share of its capital went where, when it took its first and its average at each spot — and not one separates a champion from its field. How you draft does not matter. How WELL you draft matters a great deal and is not something anyone does repeatably.

The two in-season skills are not the same skill. Adding well and starting well correlate with each other only weakly, and each survives controlling for the other. Nor is either the same as health: absences and lineup quality correlate at r=0.131, and controlling each for the other makes both links to scoring stronger (n=324).

Why nobody has run away with it. 13 coaches have won the 24 titles of 37 who ever held a roster, the most being 5. Against a league where each season's champion is drawn at random, that is unremarkable — a random league produces a 5-title coach 15% of the time. The skills are real and they are small: r ≈ 0.17 and 0.19 year over year. Real enough to matter in a season, too small to compound into a dynasty over twenty-four.

How the title team was assembled

Every champion whose championship lineup reconstructs, by where its eight starters came from.

How many of the eight who started the championship game were acquired AFTER the draft. Green is a majority of the lineup; grey is fewer than half. Across 23 champions the average is 1.6 kept, 2.6 drafted, 3.7 added.
024682002 Blackbeards0 kept · 0 drafted · 8 added converged wk 102003 Panthers0 kept · 6 drafted · 2 added converged wk 92005 Blackbeards1 kept · 4 drafted · 3 added converged wk 162006 Bickering Wrens1 kept · 2 drafted · 5 added converged wk 162007 Gamblin' Garz1 kept · 4 drafted · 3 added converged wk 162008 Send in the Clowns2 kept · 1 drafted · 5 added converged wk 162009 Blackbeards2 kept · 2 drafted · 4 added converged wk 162010 Blackbeards2 kept · 3 drafted · 3 added converged wk 152011 Bickering Wrens2 kept · 4 drafted · 2 added converged wk 122012 One Man Wolfpack2 kept · 2 drafted · 4 added converged wk 162013 South Park Cows3 kept · 2 drafted · 3 added converged wk 162014 South Park Cows3 kept · 2 drafted · 3 added converged wk 142015 Murray Creek Station2 kept · 2 drafted · 4 added converged wk 152016 California Shrimp0 kept · 4 drafted · 4 added converged wk 162017 I AM GROOT3 kept · 1 drafted · 4 added converged wk 102018 Bickering War Wrens1 kept · 2 drafted · 5 added converged wk 162019 Mr. Big Chest4 kept · 1 drafted · 3 added converged wk 162020 Man Without Fear3 kept · 3 drafted · 2 added converged wk 72021 Shut up Lige1 kept · 2 drafted · 5 added converged wk 102022 Darkwa Express1 kept · 4 drafted · 3 added converged wk 172023 Gamblin Gars2 kept · 0 drafted · 6 added converged wk 162024 Octomore 15.30 kept · 5 drafted · 3 added converged wk 172025 Rashaud Express2 kept · 4 drafted · 2 added converged wk 15

Kept, drafted, or added after the board closed — for the eight who actually started the title game, not the roster around them. 23 of 24 champions; 2004 have no reconstructable championship lineup.

Convergence is the sharper number. The median champion's title-game eight were first all on the roster together in week 16, and restricting to the 14 champions whose season reconstructs completely gives week 16 — so it is not an artifact of missing early weeks.

The extreme case is the one that started 0-4. 2023 Gamblin Gars started the title game with 0 drafted players and 6 it had picked up in-season — Kyren Williams in week 6, Isiah Pacheco in week 7, Isaiah Likely in week 16.

How to read the numbers on this page

Two different questions, two different samples, and 49 tests between them.

A champion is a bad sample and it is worth saying why. A title is a good team multiplied by a six- or seven-team single-elimination bracket — given a berth, five in six contenders lose. With 24 champions a paired test only finds an effect when the average champion sits near the 75th percentile of his own field, which almost no coaching behaviour is. So most of this page tests against all 324 team-seasons and a continuous outcome (where a team's scoring ranked within its own season), which sees an effect roughly four times smaller for no extra data. The 24 champions are then laid over that axis as a description, not used as the sample it comes from.

Two families, corrected separately. 30 champion-vs-field comparisons (sign test on paired differences) and 19 team-season correlations (Spearman, within-season z-scores so no era can carry a result). Each family gets Benjamini-Hochberg at 5% false discovery, which is the lenient correction. Champion-vs-field survivors: regPf, ceiling, missedWeeks, aboveMedian, keyPlayerMissed — and one of those is that champions outscored their field, which is nearly a definition. Team-season survivor: lineupEfficiency, acquisitionQuality, draftExecution.

Anything else on this page that looks like a finding is labelled suggestive or flat, and the facts — a bracket, a record, a seed — need no test at all.

The answers

findingWhat actually predicts success in this league?r=+0.384, p<0.001, n=324 · survives

Starting the right players. Lineup efficiency — the share of your roster's best available points that you actually put in the eight — is the strongest measure on this page and the only one of 19 team-season tests to survive correction. It predicts a team's scoring rank (r=+0.384, p<0.001, n=324) and it predicts making the playoffs (r=+0.295, p=0). This measure was reported as a null earlier in this project, because over 24 champions it is 13 of 24 and invisible. The champions are better at it than their field — mean z of 0.27 — just not by enough for 24 rows to see.

The obvious objection, tested: a team with one dominant player has an easy lineup decision, so is this just roster shape? Efficiency correlates with top-three concentration at only r=0.16, while its link to scoring is r=0.40. Concentration explains a little of it and not most of it.

findingDid champions have better injury luck?20 of 24 · p=0.002 · survives

The one comparison that survives correction. A champion lost 10.5 player-weeks to absence against 15.6 for its own season's field — about a third fewer — and was above its field in only 4 of 24 seasons. Among the players a team committed to in the first three rounds it is 2.1 against 3.5. Labelled INFERRED: the measure is a floor on disruption, not an injury count.

butDid any champion win one while losing a key player?4 of 24

Yes — 2025 Rashaud Express (Malik Nabers, 9 weeks), 2017 I AM GROOT (Odell Beckham Jr., 8 weeks), 2023 Gamblin Gars (Jeff Wilson, 6 weeks). Good availability is the tendency, not the requirement.

noAre champions built on stars, or on depth?higher in 13 of 24

Neither. Top-three concentration is 46.3% against a field 45.0%, and distinct starters differ by -0.6. Swept across every cut from top-1 to top-6, the verdict never leaves coin-flip territory — so this is the measure saying nothing, not a threshold hiding something.

maybeAre champions drafted, or forged?playoffs: 16 of 24 · p=0.152 · does not survive

Both, and the scope decides which. Over the regular season a champion's points from its own draft board are 71% against a field 71% — identical. In the playoffs it is 57% against 61%, measured against the other playoff teams, and champions shed own-draft share nearly twice as fast getting there (-14 points against -11). But it does not survive the correction — p=0.152 on its own, before accounting for the 30 comparisons this page makes. The direction is consistent and the mechanism is plausible; the evidence is 21 seasons and that is not enough. Treat it as the most interesting thing here to go and test properly, not as something established.

noDid champions trade more?higher in 14 of 24

No signal — 1.7 recorded trades against 1.3, and waiver activity is the same coin flip (14 of 24). Read as a floor. The CBS transaction log is incomplete — the roster replay it drives cannot hold rosters at fourteen — so these counts are what was recorded, not what was done.

noWas there a pivot point — a week it turned around?0 of 24

Not one. Every season has a best split; the question is whether it beats the season's own variance, tested here by shuffling each team's weeks two thousand times. No champion's season splits more sharply than chance. Across all 324 team-seasons in the archive the same test fires on 13 — 4.0% at α=0.05, which is what a calibrated test does. Champions were not teams that turned a corner. They were mostly good all year, or good enough and lucky in January.

flatDoes the draft slot that wins exist?r=-0.006, p=0.916, n=316

No. It is the flattest measure in the whole study. Draft slot against a team's scoring rank across 316 team-seasons: r=-0.006, p=0.916, n=316. There is no lucky chair.

It would be a misleading number even if it were not flat: this league assigns the order inverse to finish and the champion picks last the following year, so a champion's slot is a statement about the season before it. What matters in a keeper league is your first LIVE pick, which is your slot and everyone else's keeper count together.

flatDid champions draft differently — positions, experience, rookies, keepers?every input tested, none survives

No, and this is the most comprehensively negative result here. Each of these is a decision made on draft night, before a down is played, tested across 324 team-seasons against how that team then scored:
· when they took a quarterback — r=-0.013, p=0.819, n=288
· when they took a running back — r=+0.023, p=0.671, n=311, a receiver — r=+0.058, p=0.304, n=310
· keepers held — r=+0.112, p=0.074, n=292
· NFL experience of the players they drafted — r=-0.120, p=0.051, n=316
· rookies drafted — r=-0.063, p=0.258, n=316
· NFL draft pedigree of the players they took — r=-0.047, p=0.386, n=316
Not one survives correction. The two that come closest point the same way and both are weak: taking a tight end later (r=+0.179, p=0.042, n=159) and buying fewer late-round lottery tickets (r=-0.122, p=0.048, n=316) — and the second is more likely a symptom of already having a good roster than a cause of getting one.

departureWas the title year that coach doing what he always does, or something different?22 of 24 · p&lt;0.001 · within-coach

A departure, and it is the strongest result on this page. Comparing a champion to its league says whether the team was good; it cannot say whether the COACH was doing anything unusual, because a coach who is always good beats the field every year. Holding the coach fixed and comparing the title season against his own other seasons, 22 of 24 scored at a higher weekly rank than their norm, by 8.97 percentile points. Measured in percentile rather than points on purpose: on raw points it is only 16 of 24, because the scoring rules changed four times and a coach whose title came late looks improved when the rulebook did.

But only 3 of 24 were that coach's best season ever. The title year was above their line and short of their peak — which is the most practical thing here: you do not need a career year to win this league.

noDid champions manage their lineups better?13 of 24 · p=0.839

No, and of all the nulls this is the one worth sitting with, because it is the one that would have been skill. Points actually started as a share of the best eight the roster held that week: 78.5% for champions against 77.0% for their field. They did not start the right players more often than the teams they beat. Nor were they steadier — weekly consistency is 12 of 24, p=1.000 — and their bad weeks were no better than anyone's (floor, 15 of 24). What was different is the top end: their best week beat their field's in 21 of 24.

noDid champions lean on one position?RB: 16 of 24 · p=0.152

No — and this page said the opposite first time round, which is worth recording. Running back looks like a lean: 29% of a champion's starter points against a field 24%. But it is 16 of 24 seasons at p=0.152. The first version quoted the well-covered subset instead (13 of 15) and called it the one signal that holds — choosing the slice where the number looks best, out of an already non-significant result. No position separates them.

yesCan a team start 0-2 and still win it?6 of 24 did

6 champions started 0-2 and 2 started 0-4. The strongest case is 2023 Gamblin Gars, who started 0-4 head-to-head and 0-4 against the league median — 0-8 by the standings this league actually keeps — and won the title. Three champions also came in as the last team in the field: 2012 at the 7 seed, 2014 at the 7 seed, 2024 at the 6 seed.

What actually predicts success

Every input and output measured here, against how a team scored, over 324 team-seasons.

How strongly each thing a team did relates to how it then scored, across 324 team-seasons — not 24 champions. Bars are the absolute correlation; the sign and the sample are beside each. Green is the one measure that survives correction for having run 19 tests. Everything below it is grey because it did not.
0.000.100.200.300.40starting the right eight+0.384 n=324 ✓acquisitionQuality+0.365 n=309 ✓draftExecution+0.343 n=316 ✓keeperSurplus+0.246 n=150took a TE later+0.179 n=159player-weeks lost to absence−0.153 n=324late-round lottery tickets−0.122 n=316experience drafted−0.120 n=316keepers held+0.112 n=292age of players drafted−0.103 n=316took a kicker later+0.099 n=269took a defence later+0.079 n=314rookies drafted−0.063 n=316took a WR later+0.058 n=310NFL draft pedigree−0.047 n=316took a RB later+0.023 n=311took a QB later−0.013 n=288draft slot−0.006 n=316experience of kept players−0.005 n=243

Read the bottom of that chart, not the top. Everything from keepers held downward is a decision made on draft night, and none of it predicts anything. The draft is where this league spends its attention and it is not where the league is decided.

Lineup efficiency contains outcome and is labelled as decision quality, not pure behaviour — it is computed from what the players actually scored, so a coach cannot be graded on it in advance. It is also an upper bound: the best eight here is the best eight by points, ignoring slot eligibility, so every team's figure is inflated by the same rule and it is the ordering that is being compared, not the level. The legal version is solved on the near misses page.

How they drafted, position by position

Four dimensions per position, because no one of them is the answer on its own.

"Took a receiver late" is not a measurement. A team that drafted one receiver in round 2 and a team that drafted one in round 2 and five more in rounds 9-14 have the same "first WR" pick and nothing else in common; a team that waited until round 5 and then took four straight looks late at the position it committed to hardest. So each position carries how many they took, what share of their draft capital went there, where the first one went and where the average one went — and capital is the one that weighs round against quantity, because it prices every pick off this league's own curve.

How many they drafted
poschampionfieldzp
QB1.71.50.220.40
RB2.82.9-0.070.78
WR3.53.50.060.72
TE0.70.8-0.200.36
K1.00.90.350.20
DEF1.31.3-0.030.90
Share of draft capital spent there — the one that weighs round against quantity
poschampionfieldzp
QB14%12%0.290.22
RB25%26%-0.090.73
WR32%30%0.200.35
TE5%7%-0.280.19
K7%6%0.140.55
DEF11%11%-0.030.88
Where the FIRST one went, as a percentile of that season's live board (0% = first pick)
poschampionfieldzp
QB32%40%-0.290.16
RB24%19%0.220.37
WR21%20%0.070.72
TE39%42%——
K86%78%0.380.06
DEF56%55%0.140.55
Where the AVERAGE one went — this is what separates "early and few" from "late and many"
poschampionfieldzp
QB46%52%-0.330.22
RB45%41%0.210.31
WR48%44%0.260.21
TE56%51%——
K86%78%0.420.05
DEF61%61%0.040.86

22 cells, and not one survives correction. Timing is a percentile of that season's live board — 0% is the first live pick, 100% the last — so keeper-consumed rounds do not distort it and a 12-round 2005 draft compares with a 14-round 2019 one. Capital is priced from a curve fitted on 2861 skill-position picks across every season: the top 5% of a board returns 1.276× the average drafted player and the bottom 5% returns 0.507×.

The one coherent pattern, and it still does not clear the bar: champions drafted their kicker later and took fewer of them — later first pick (z=+0.38), later on average (z=+0.42, p=0.046), fewer taken (z=+0.35). Three dimensions pointing the same way is not what one lucky cell looks like. A kicker is the only asset on the board whose value next August is guaranteed to be zero, so spending early there is a pure vote for this November — and champions voted less.

And running back, specifically: champions spent no more capital there than their field (z=-0.09, p=0.73) and drafted no more of them. An earlier version of this page reported that champions took more of their POINTS from running backs. That is an output. There is no decision behind it.

Availability is the separation, among champions

Of everything measured here, this is the one that is not close.

Player-weeks lost to absence — champion against its own season's field. Green is the champion; grey is the mean of the other teams that season. Lower is healthier. 24 seasons.
championrest of that season08152330200242003320049200552006820071220081420096201062011132012620133201410201572016102017172018112019102020202021820228202326202415202522

A missed week is a player on the roster whose NFL team played and who has no stat line. It is a floor on disruption, not an injury count — it catches suspensions and healthy scratches too, and every figure from it is INFERRED. The status verbs that would not have that problem do not cover the era: the CBS log carries 13 Moved to IR events in 11,134.

Drafted, or forged? Both — ask again in January

Over a regular season a champion's roster origin is indistinguishable from anyone else's. The playoffs are a different team.

Share of a champion's starter points that came from players it drafted or kept itself, regular season against playoffs. One faint line per champion, 24 of them. The thick green line is the champion mean; the orange line is the mean for the other playoff teams. Both fall — a January roster is always less the draft than an October one — but champions fall nearly twice as far (-14 points against -11).
regular seasonplayoffs0%25%50%75%100%other playoff teamschampions

Origin is split three ways: players the team drafted or kept itself, players another team drafted that season and it acquired later, and players nobody drafted. The playoff comparison runs against the other playoff teams only — comparing a champion's postseason against teams that had none would be comparing it against zero. The most forged title teams on record: 2002 Blackbeards, 100% of its playoff points from players nobody drafted; 2022 Darkwa Express, 40% of its playoff points from players nobody drafted.

The title year against the coach's own career

The one comparison that holds the coach fixed — and the one that separates hardest.

Mean weekly scoring percentile in the title season (green) against the same coach's average across every other season he has played (grey). 22 of 24 champions are to the right of their own norm. Only 3 were career bests.
championrest of that season20385573902007 Scott Garlisch752013 Brian Pugh652017 Kevin Herndon702006 Mike Pugh712025 Todd Mercer632023 Scott Garlisch582002 Terry Herndon652003 Jimmy OBrien642004 Ben Byler532015 Ryan Ayers652005 Terry Herndon642021 Terry Herndon632016 Dan Weissburg532022 Todd Mercer592009 Terry Herndon622011 Mike Pugh622020 Matt Pugh572010 Terry Herndon592008 Steve Baird602014 Brian Pugh492024 John Silvers512019 Kevin Herndon562012 Kevin Herndon512018 Mike Pugh51

Each row is a champion. The dot is that season's mean weekly scoring percentile; the line runs back to the average of every other season that coach has ever played, across team renames and both platforms. 22 of 24 moved up, by 8.97 points on average. Percentile, not points, because points are not comparable across a career here — the scoring changed four times, and on raw points the same test is only 16 of 24.

Titles are concentrated. 13 coaches have won the 24, out of 37 who have ever held a roster — 35% of the league's history has a ring. Terry Herndon has 5 (2002, 2005, 2009, 2010, 2021); Mike Pugh has 3 (2006, 2011, 2018); Kevin Herndon has 3 (2012, 2017, 2019).

The shape of a championship season

Weekly scoring percentile within the league, 24 champions. Green dots are wins, red losses. The flat line is the league median for that week.

200211-3
200310-4
20049-5
20058-5
20069-4
200711-2
20086-7
20098-5
20108-5
20118-5
20126-7
201311-2
20145-8
20158-5
20169-4
20177-6
20189-4
20197-6
20206-7
20219-5
202210-4
20237-7
20244-10
202510-4

Not one of these is a pivot. Every series has a best split into a worse half and a better half — that is arithmetic. Whether it means anything is tested by shuffling each champion's own weeks two thousand times and asking how often chance produces a split that sharp. It produces one at least that sharp for every single champion. The same test fires on 13 of the archive's 324 team-seasons, so it is calibrated and it is not broken; champions simply did not turn corners.

The worst starts, and the teams living one now

Wins in the first 2 weeks, counted the way this league's standings count them — head-to-head and the median game, so 4 decisions rather than 2. Green: the eight worst starts by an eventual champion. Grey: the 2026 teams still without a head-to-head win. Both sides are measured over the same 2 weeks, because the season in progress has played 2.
012342014 South Park Cows0-2 · 0-2 med2016 California Shrimp0-2 · 0-2 med2023 Gamblin Gars0-2 · 0-2 med2012 One Man Wolfpack0-2 · 1-1 med2019 Mr. Big Chest0-2 · 1-1 med2004 Bulldogs1-1 · 1-1 med2005 Blackbeards1-1 · 1-1 med2008 Send in the Clowns1-1 · 1-0 med2026 Rashaud Express0-2 · 1-1 med — now2026 Calabash Shrimp0-2 · 0-2 med — now2026 Octomore 15.30-2 · 0-2 med — now

As of 2026 week 2, Rashaud Express is 0-2 and 1-3 including the median game, Calabash Shrimp is 0-2 and 0-4 including the median game, Octomore 15.3 is 0-2 and 0-4 including the median game. The precedent is real and it is specific: 2023 Gamblin Gars started 0-4 and 0-4 against the median — 0-8 by the standings this league keeps, over its first four weeks — and won the title from the 5 seed. Over the full four weeks 2 champions were 0-4 head-to-head and 5 were below .500. What that team then did is the part worth copying: 38% of its playoff points and 11 of its 24 playoff starts came from players nobody in the room drafted.

Running back, and nothing else

Share of starter points from running backs, champion against its own season's field. The one positional lean that survives the comparison.
championrest of that season10%19%28%36%45%200239%200331%200414%200523%200618%200715%200823%200928%201029%201123%201218%201333%201434%201538%201636%201736%201844%201928%202020%202128%202239%202326%202435%202536%

The twenty-four

Sortable. own draft is the share of starter points from players the team drafted or kept itself; weeks lost counts roster players whose NFL team played and who posted no stat line.

seasonteamcoachseedrecordpct keptallowedtop-3own draft regown draft po weeks losttradesmargin
2002BlackbeardsTerry Herndon111-30.786——41%0%0%4110
2003PanthersJimmy OBrien110-40.7140044%92%61%304
2004BulldogsBen Byler49-50.6430041%57%69%9026
2005BlackbeardsTerry Herndon38-50.6152246%75%56%5437
2006Bickering WrensMike Pugh19-40.6921236%51%40%845
2007Gamblin' GarzScott Garlisch111-20.8462253%79%64%12233
2008Send in the ClownsSteve Baird56-70.4623438%52%19%14224
2009BlackbeardsTerry Herndon28-50.6154443%62%55%621
2010BlackbeardsTerry Herndon48-50.6154438%75%48%6146
2011Bickering WrensMike Pugh38-50.6153453%86%75%13071
2012One Man WolfpackKevin Herndon76-70.4624456%88%74%6037
2013South Park CowsBrian Pugh111-20.8464455%85%79%3332
2014South Park CowsBrian Pugh75-80.3854451%74%78%10219
2015Murray Creek StationRyan Ayers28-50.6154443%71%54%7314
2016California ShrimpDan Weissburg49-40.6920449%87%55%1011
2017I AM GROOTKevin Herndon47-60.5384441%73%54%17348
2018Bickering War WrensMike Pugh29-40.6922455%51%33%11361
2019Mr. Big ChestKevin Herndon67-60.5384443%71%75%10027
2020Man Without FearMatt Pugh56-70.4623451%84%72%20237.4
2021Shut up LigeTerry Herndon19-50.6433447%68%53%8240.9
2022Darkwa ExpressTodd Mercer210-40.7141447%95%57%8039
2023Gamblin GarsScott Garlisch57-70.5004445%62%48%26376
2024Octomore 15.3John Silvers64-100.2860443%83%73%15229.15
2025Rashaud ExpressTodd Mercer310-40.7143451%81%83%22178.35

What this can and cannot see

The population is two numbers. 24 champions for anything computed from games; 24 for anything computed from players. Playoff figures cover 24: 2004 and 2007's bracket lineups do not reconstruct.

Lineup coverage is uneven and the weak seasons are named. A CBS team-week contributes player rows only if its eight starters reconstruct to the recorded score. Field coverage is 100% in every Sleeper season and falls to 55% in 2020 and 63% in 2010. 9 seasons sit below 90% (2002, 2003, 2004, 2005, 2006, 2007, 2008, 2010, 2020), and every aggregate above is computed twice — over all seasons, and over only the well-covered ones. A finding that moved between the two is not reported as a finding.

Trade and waiver counts are floors. The CBS transaction log under-records departures; the roster replay it drives leaves 1,621 of 4,182 roster-weeks with too many players.

Thresholds are swept, not asserted. Top-N concentration is reported across N=1..6 and never separates; the key-player cut is reported across rounds 1..6 and the champion advantage deepens monotonically, so the headline understates it.