Four ratings, computed independently, each answering a different question about the same season. None of them is the ranking — the point of showing all four is that they disagree, and where they disagree is usually the interesting part.
What does the money think?
Betting-market rating — market-implied points vs. league average, from closing spreads (ridge regression, HFA-adjusted).
How it's built: Closing point spreads solved into weekly ratings (ridge, HFA-adjusted).
What does the win-loss history predict?
A running team-strength rating (the classic Elo system): 1500 is average, and each game moves points from loser to winner — more for an upset or a blowout. Ratings carry across seasons, reverting part-way toward average each offseason.
How it's built: Self-computed 538-style Elo over every game since 2000. Learns only from results, so it prices in a mid-week injury slowly — that lag is why Market stays in the blend.
How well are they actually playing?
Net expected points added per play, offense and defense combined. Early in a season this is pulled toward league average until a team has played enough games to trust the sample.
How it's built: Opponent-adjusted net EPA per play from play-by-play (schedule-adjusted).
What have they earned?
Opponent-adjusted win-loss — win percentage weighted by the strength of the teams each result came against. Beating good teams counts more than beating bad ones. Early in a season it's pulled toward league average until a team has played enough games to trust the sample.
How it's built: Margin-blind: opponent-adjusted wins and losses only, scores discarded.
The four ratings are in four different units — points, Elo, EPA per play and a win percentage — so they can't be averaged as they are. Each week, every lens is standardized across the 32 teams into a z-score: how many standard deviations from league average that team sits, on that lens, that week.
Cell colour on the board comes from that z-score. The displayed number stays in the lens's own native unit, which is why two cells can share a colour and show very different numbers.
Efficiency and Meritare noisy over a handful of games, so early in a season each team's z-score is pulled toward league average by how much of a sample it has: the surviving share is games / (games + 4). At one game roughly a fifth of the signal survives; by ten games, about seven-tenths.
Market and Elo are left alone — Market self-regresses through its own preseason prior, and Elo carries across seasons, so neither starts from nothing.
Weighted combination of Market, Elo, Efficiency and Merit, in standard deviations from league average.
The default weighting is Market 40%, Elo 25%, Efficiency 20%, Merit 15%. It is deliberately not equal-weight: Market and Elo are both forward-looking and move together, so weighting them equally with the rest would count that one axis twice.
These are relative weights, not a budget — they're renormalized over whichever lenses a season actually has, so a slider at 100 doesn't mean 100%.
Spread (standard deviation) of the four lenses' z-scores for a team. High = the lenses tell different stories about them.
Divergence is the spread among the four lenses. In practice it mostly flags teams the market values differently than their results, since Elo, Efficiency and Merit tend to agree by season's end, so the money is usually the odd one out.
Ranking by Divergence does not rank strength. #1 on that tab is the team the four lenses argue about most, which is frequently a losing team the market still rates. It also ignores your blend weights entirely — it measures the spread between the lenses, not any combination of them.
Market and Elo correlate ~0.85 (both forward-looking on team strength); the three results lenses (Elo, Efficiency, Merit) correlate ~0.90+ by season's end.
So “four independent lenses” overstates it. Three of the four are results-based and converge as a season fills in; Market is the one that routinely disagrees, and Merit is genuinely orthogonal in that it throws away margin entirely. We show all four because they answer questions a reader actually asks separately — not because they're statistically independent.
Every lens is regular-season only. Nothing here is a prediction of a game, a playoff seed, or a betting recommendation.
Play-by-play, schedules, results and closing lines come from nflverse, licensed CC BY 4.0. Boards are recomputed weekly in season; the board header carries the week it reflects and, where the publisher stamped one, the date it was last recomputed.