League Intelligence

All-play record

Each week, every team is replayed against all eleven others. With twelve teams a week produces eleven results instead of one, which is why all-play settles down far faster than a win-loss record does. Tied scores count as half a win.

allPlayPct = (wins + ties / 2) / games

Expected wins and luck

A week’s all-play share is also the chance of beating an opponent drawn at random from the league, so summing it gives the wins a team would expect from a neutral schedule. Luck is the difference. Summed over a season the weekly figures telescope into exactly this number, so the two can never disagree.

weeklyLuck = outcome − expectedWinPct   (win 1, tie 0.5, loss 0)
seasonLuck = actualWins − expectedWins = Σ weeklyLuck

Median record

The league’s middle score each week. With twelve teams there is no single middle team, so the median is the midpoint of the 6th and 7th scores — which means a team can only tie it when those two are themselves equal. An exact tie counts as half a win.

medianWin = score > median ? 1 : score < median ? 0 : 0.5

Opponent percentile

How much of the league your opponent outscored in the week you played them. The team being measured is left out of that comparison: if it were included, scoring badly yourself would raise your opponent’s percentile and your schedule would look hard precisely because you played badly.

opponentPercentile = (teams the opponent beat + ties / 2) / (teams − 2)

Power Index

All-play at 45%, scoring at 35% and recent form at 20%, each on a 0-100 scale. Scoring is scaled between the league’s weakest and strongest average rather than into standard deviations, because a standard deviation estimated from a handful of weeks is unstable enough that one big week would redraw the table. Recent form is left out until three weeks exist and its weight is shared between the other two in proportion.

index = 0.45 × allPlay + 0.35 × scoring + 0.20 × form

Positional strength

Starter points grouped by the player’s own position rather than by the slot he filled. A running back started in the flex scored his points as a running back; filing them under "FLEX" would make every running-back total wrong and would treat a hole in the lineup as though it were a position. Bench points are excluded, and a position a team has never started shows an em dash rather than a zero.

Games cost by a lineup decision

A loss where the best legal lineup from that week’s roster would have outscored the opponent’s submitted score. The comparison is deliberately one-sided: the opponent played the lineup they played, and imagining they had also been perfect would answer a question nobody asked. Losses no legal lineup could have won are counted separately, because the two are not the same failure.

gamesCost = losses where optimalPoints > opponentScore

What this is not

None of it is a prediction. The Power Index describes games already played; it says nothing about next week, and no component is fitted, tuned or learned. Yahoo’s standings rank and each team’s record are deliberately excluded from every figure on this page except the columns explicitly labelled "actual", which are there for contrast.