Methodology
The inputs are public. The weights are not. Here is everything in between.
The Promise
Every ranking on this site is published as a dated post. The change column is computed against the prior week's file, not typed, which means it cannot be quietly adjusted to look smarter in hindsight. If a call was wrong, the archive says so. Corrections get their own box; the original stays where it was.
That is the whole idea. Plenty of people publish rankings, and some will tell you about the biggest risers and fallers. Nobody shows the movement on every player and a full list, every week, computed rather than typed, with the model's rank and the editor's rank side by side and last week's file still there to check. That is the gap (at least to my knowledge). While weekly rankings are paramount to winning your week, you are ultimately building a team for a whole season, and a championship.
How a Rank Gets Made
The model starts with the team, not the player. Each of the 32 offenses gets a projected season total for rushing attempts, pass attempts, yards and touchdowns, seeded from Vegas win totals and last season's rates, regressed toward the league. Those totals are the ceiling. No amount of optimism about a player can add carries the offense doesn't have.
Each player is then assigned a share of his team's totals: his slice of the carries, his slice of the targets. A share, multiplied by a team total, becomes a stat line. A stat line, scored in half-PPR, becomes points per game. Points per game becomes the rank.
The shares are where judgment lives, and the model is honest about that. A returning starter inherits last season's role by default. A player on a new team, or a rookie, has no default worth trusting, so those are set by hand, and the write-up says so.
What the Metrics Do
Four proprietary metrics carry the football rankings. None of them touches the rank directly. Each one multiplies a specific stat, which is a deliberate constraint: because every metric acts on a named input, every rank change traces back to a number, and the number traces back to a reason. That is the sentence in the movers sidebar every week.
OTS — Opportunity Trend Score
The recent trend in a player's share of his team's opportunity, normalized across the league; it scales projected volume. Opportunity means targets for receivers and tight ends, carries plus targets for running backs. A rising OTS means the role is growing before the box score shows it. OTS scales projected volume, and it works at every skill position, which makes it the spine of the system.
RCS — Role Clarity Score
A running back's share of the touches across early downs, passing downs, the red zone and the two-minute drill, scored 0 to 100; it scales projected rushing touchdowns. Volume metrics tell you how much a back is used. RCS tells you what kind of back the team thinks it has. A back who owns early downs but disappears on third down is a different asset from one who stays on the field in all four situations. The score is labeled Workhorse, Feature Back, Role Player or Specialist, and it scales projected rushing touchdowns because goal-line trust is what the label is really measuring.
SFS — Scheme Fit Score
How a receiver's own profile, his share of air yards, depth of target and yards after the catch, fits the tendencies of the offense he plays in; it scales projected receiving yards. Receivers and tight ends. It matters most after a coordinator change or a trade, which is exactly when rankings move the furthest and explanations are thinnest. SFS scales projected receiving yards.
PGHI — Passing Game Health Index
A composite of a quarterback's passing environment: pass rate over expectation, the protection trend, whether the weapons are on the field and how fast the offense plays; it scales projected passing touchdowns. Quarterbacks. It answers whether things are getting better or worse for the quarterback before his stat line reflects it. PGHI scales projected passing touchdowns.
What Stays Private
The inputs above are public. Every one of them is built from play-by-play and roster data that anyone can download. What is not public is how they are weighted: the coefficients, the regression targets, how much each metric is trusted in September versus November. That is the work, and it stays on this side of the page.
For the record, this is not an effort for secrecy. It is the difference between telling you what the model looks at and handing you the model. The first is what makes a ranking worth trusting. The second is what makes it worth copying.
Where Judgment Comes In
The model produces a rank for every player. Then I move the ones I disagree with, on a board, by hand, and both numbers publish. The rankings table shows where the model had him and where I put him, so every override is visible as a gap between two columns. A move of more than a few spots needs a written reason before the file will export; where a note is still missing, the table shows a dash, not a guess.
The annual Flag Plant is the purest version of this. It is a rookie call made on film and instinct, published before the season, graded honestly after it. The model has an opinion. So do I. When they disagree, you see both.
Tiers and Formats
Tiers are drawn where the points-per-game cliffs actually are, not at even intervals. The gap between tiers is the useful information; the tier number is just a label.
Rankings default to half-PPR redraft. Superflex, dynasty and points-league variants are planned.
Questions
What is half-PPR scoring?
Half a point for every catch, on top of the usual yards and touchdowns. So 0.1 per rushing or receiving yard, 6 per touchdown, 0.5 per reception; for quarterbacks it's 0.04 per passing yard, 4 per passing touchdown and minus 2 per pick. Every rank on this site is built on that scoring. If your league is full PPR or standard, pass-catching backs and possession receivers will sit a little higher or lower for you, and that's fine, the order mostly holds.
What does rest of season mean?
It means I'm ranking who helps you most from this week through the end of the regular season, not just this Sunday. That's the question you're actually answering when you make a trade or put in a waiver claim. Weekly rankings are great for setting a lineup. These are for building a team.
What does the change column compare against?
Last week's file. Week 1 compares against the preseason baseline, and every week after that compares against the week before. The number is computed from the two published files, never typed, so I can't quietly nudge it later to look smarter (tempting as that'll be some weeks). The preseason edition is the baseline, so its change column is a dash.
What is the Role Clarity Score?
It's my running back metric, and it asks a simple question: what kind of back does the team think it has? It measures his share of the work across early downs, passing downs, the red zone and the two-minute drill, on a 0 to 100 scale. A back who's on the field in all four situations is a different asset from one who only sees early downs, even if their carry counts look the same, and the score labels him accordingly: Workhorse, Feature Back, Role Player or Specialist. It scales his projected rushing touchdowns, because goal-line trust is really what the label is measuring.
Why is there a model rank and a board rank?
Because I want you to see where I overrode the math. The model rank is where the projection put him on team totals and his share of them alone. The board rank is where I put him after moving the players I saw differently, with a written reason for anything more than a few spots. Both numbers publish on every row. When they match, I trusted the projection. When they don't, the gap between the two columns is me, and the note says why.
How often do the rankings update?
Once a week during the season, Tuesday morning, in time for waivers. Each edition gets its own dated page that never changes, the previous week's file stays exactly where it was, and the archive lists every one back to the preseason baseline. If something big happens midweek I'll say so on X, but the rankings themselves move on Tuesdays.
What stays private?
The weights. Every input is public: the metrics are built from play-by-play and roster data anyone can download, and this page tells you what each one measures. What I keep is how much each metric moves a stat, the regression targets, and how much I trust the metrics in September versus November. That's the difference between telling you what the model looks at and handing you the model. The first is what makes a ranking worth trusting; the second is what makes it worth copying.