The Rankings FAQ: How to Read Our Board

What Each Column Is Doing

Every row on our projection board is one player and four separate claims about him. The median is our middle outcome for the season, the band is the spread around that median, the rank orders players by median inside a position, and ADP is what the market currently charges to get him. Each column does separate work, and a manager reading one of them is drafting off a quarter of the page.

The median, the P50, is the point where half of our simulated seasons land above it and half land below. It is the number the ranking sorts on, and it is the number to compare against another player's median when you are choosing between two rows. Compare medians against medians; that is the comparison the column was built for.

The band is the tenth and ninetieth percentile of that same distribution. A narrow band says the projection rests on usage we have already watched, repeated, and measured. A wide band says most of the outcome depends on a role nobody has assigned yet: a camp battle, a snap share that could double or vanish, a target tree with three plausible shapes. The inputs behind those distributions get their own page; this one stays on the output.

Rank is a derived column. Positional rank falls out of the median ordering, and the overall board weighs each median against replacement level at that position, which is why a tight end and a running back projected for identical points can sit fifty picks apart. That weighting argument is made in full on the value-based drafting page. Scoring format moves the medians themselves, so the PPR board is a genuinely different board from the standard one.

Why A Rank Disagrees With ADP

Two processes produce those two numbers and they run on different fuel. The model prices role: snap share, target share, carry share, red-zone opportunity, and the offensive environment those shares sit inside. Draft position prices the room, which runs on name recognition, last season's finish, and whichever camp report moved through the timeline this week.

Agreement is the common case and the boring one. On any given run, most rows sit within a few picks of their price, the gap reads near zero, and there is nothing there to act on. The rows worth your attention are the ones where the two processes landed far apart, because one of them is wrong and the size of the gap measures how much rides on which.

A wide gap is a hypothesis we are willing to be graded on. Some of those hypotheses lose: a projection can call a backfield split correctly, miss the touchdown rate badly, and finish the season under its price anyway. The accuracy ledger keeps the losses in the same table as the wins at the same precision, which is the only version of that table worth publishing. The gap-hunting piece works the value side of the column in depth; the point that belongs here is smaller. A disagreement is information about a disagreement, and the season is what settles it.

What The Band Changes About Your Pick

Identical medians can sit on completely different distributions, and the band is the column that says so. A veteran back with three seasons of stable workload might carry forty points between his tenth and ninetieth percentile. A rookie receiver behind two established starters can carry a hundred-point spread around the same median, because his season contains both a redshirt year and a weekly starter.

Roster construction is where that distinction pays. Early picks are where you buy narrow bands, because a floor collapse in the first three rounds costs you the season while an unreached ceiling costs you a good bench. Late picks invert the logic: a wide band at pick 140 costs one roster spot to hold and returns a starter whenever the top of the distribution lands.

A projection you disagree with is easier to argue with once you read its band. Width tells you whether the model is confident and wrong or uncertain and wrong, and those two are different conversations to have with your draft plan.

A Stretch Of Board, Worked By Hand

Take an invented stretch of receiver board: six rows, numbers made up to show the shape, not rows from any current run.

The top three rows sit close: medians of 214, 209, and 206 points, bands of 178 to 249, 165 to 256, and 190 to 223, prices of 18, 24, and 41. The bottom three fall away. A 181 median (band 141 to 228) is priced at 33, a 178 (160 to 197) at 52, and a 176 with the stretch's widest band (120 to 241) at 46.

The spacing between medians goes 5, 3, 25, 3, 2. The 25 clears its surrounding spacing eightfold, and the other four gaps sit inside what we treat as the noise of a season-long projection. Where noise ends and a real break begins is a judgment cutoff, flagged as one, and the multiple-of-local-spacing test behind it puts the break after the third row: an upper tier of three players.

The prices tell a different story about those same three rows. They read 18, 24, and 41, which leaves the third row seventeen picks behind the second one while its median trails by three points. That seventeen is the whole actionable content of the stretch.

Say your turn is pick 30 and your next is pick 39. The first two rows are gone before you pick at all. The third row is priced at 41, two picks past your second turn, so the arithmetic says you can spend pick 30 somewhere else and still have a live shot at the tier when you come back around. ADP is an average, which means a price of 41 goes at 32 in some rooms and 50 in others, and the cushion is a probability you are choosing to accept with your eyes open.

The fourth row is the trap in the stretch. Its median sits 25 points under the tier and its price sits eight picks ahead of the row directly above it in projection — the exact shape of a room paying for a box score that already happened.

The sixth row is the band lesson. Its ninetieth percentile of 241 lands above the top row's median of 214, which is what a wide band means and exactly why the ranking sorts on the P50: that ceiling shows up in a thin slice of simulated seasons, and pick 46 buys the entire distribution, floor included. Finding the break, counting the rows above it, and checking that count against your turn gap is the operation. The tier-break piece takes the operation apart in full; this stretch is one pass through it.

When The Board Moves Under You

The model reruns on demand. A rerun happens when an input changes enough to move a projection (a depth chart revision, a role change that shifts a projected share, a correction to the usage data feeding the distribution), and what comes out carries new medians, new bands, and new ranks together.

ADP updates daily, which means the gap column can widen or close overnight with the projection untouched. A gap that grew while you slept usually means the room repriced a player and our number held its ground. Reading a gap without knowing which side moved is how managers talk themselves into chasing the market.

This page is an instruction sheet, and the decisions get made on the surface it describes: the medians, the bands, and today's prices all live on the live board, where every number moves under the prose. Every figure in the worked example above was invented to show you a shape. Before leaning on any of it, read the accuracy record, find the calls that went wrong, and set your confidence at whatever that record earns — including your confidence in this page.