High-Variance and Low-Variance Players, Defined

Every player on a projection board carries a median and a band. The median is the number that gets read; the band is the number that decides drafts. Our boards express that band as a P10 and a P90 — the model's tenth- and ninetieth-percentile outcomes for the season — and the distance between those two numbers is what high variance and low variance actually mean here.

Width is a statement about the model's information. A wide band says the inputs behind a player disagree with each other, or that the sample they come from is thin. A narrow band says several independent inputs point at roughly the same season.

Where the line sits is a judgment call, so here it is stated as one: we read a band as wide when it spans more than half the player's own median, narrow when it spans under a third. No backtest fixed those fractions — they are working cutoffs, and the invented stretch below lands rows on both sides of them.

Two players can share a median to the decimal and be entirely different draft objects. Drafting the median treats them as interchangeable; drafting the band treats them as two bets with two different jobs on a roster. How the width gets made, and how to spend it, is the rest of this piece.

Four Things That Widen A Band

Role dispersion does most of the work. A back with a locked three-down job has a snap share that moves a few points week to week; a committee back's share can swing twenty points on game scripts nobody can see in August. The projection has to cover both worlds, and covering both worlds is what a wide band looks like from the outside.

Games-played risk is the second input, and it is the one readers most often mistake for pessimism. An age-curve flag or a multi-season soft-tissue history barely moves a median. It moves the P10 hard, because the bottom decile of that season includes the version where the player appears in nine games. Width from this source is asymmetric: the floor drops and the ceiling holds.

Efficiency dependence is the third. Production built on volume — targets, carries, routes run — repeats year over year more reliably than production built on touchdown rate or yards per touch, so a season that leaned on scoring rate generates a wider band than a season that leaned on usage. A wide band sitting on an expensive price is the shape behind most fade calls; the bust and fade argument gets its own page and its own screen.

Sample thinness is the fourth. Rookies, position switches, and second-year players with one usable season give the model fewer observations, so the prior carries more of the projection, and priors are wide by construction. How those priors are built and where they are weakest is documented on the model methodology page.

Scoring format resets part of this before a draft starts. Reception volume is among the stickiest inputs on the board, so full PPR narrows the bands on pass-catching backs and slot receivers, while standard scoring pushes those same players back onto touchdown rate and widens them. Format math is settled on the PPR, half-PPR, and standard breakdown; assume it here.

Reading Band Width On An Invented Stretch Of Board

Take an invented stretch of receiver board — the numbers below are made up to show the shape, and no run of our model produced them. Four receivers sit next to each other, priced within a round of one another, projected in full-PPR points for the season.

Receiver A projects a median of 198, with a P10 of 172 and a P90 of 226. Receiver B medians at 194, P10 141, P90 271. Receiver C medians at 191, P10 168, P90 217. Receiver D medians at 186, P10 132, P90 268.

Start with the medians. Top to bottom, the four are separated by 12 points across a 17-game season, about seven-tenths of a point per week. A board sorted on the median column presents that stretch as a smooth, nearly flat descent.

The width column reads differently. A spans 54 points, B spans 130, C spans 49, D spans 136 — the widest row is nearly three times the narrowest, on players the median column and the room both called equivalent. Against the cutoffs above, B and D read wide (each band spans two-thirds of its median or more) while A and C read narrow (just over a quarter). That split is the signal the median column cannot show.

The ordering flips depending on which end of the band you read. Sorted by P10, the stretch is A at 172, C at 168, B at 141, D at 132, and the distance from best floor to worst is 40 points. Sorted by P90 it runs B at 271, D at 268, A at 226, C at 217, and the distance from best ceiling to worst is 54. Same four rows, three different drafts.

When To Buy Width And When To Buy Floor

The rule is roster-relative. Width earns its price when a roster already clears the playoff line without it, since a median outcome that already gets you there makes the ninetieth percentile the only draw that changes the season. In that position, receivers B and D from the stretch above are the only two rows worth the pick.

Narrow bands earn their price when a single collapse ends the year. Early rounds, shallow benches, and a thin wire all convert one P10 outcome into an unrecoverable start, and A and C are what survive that. The floor difference in the invented stretch runs 40 points, a bit over two a week, and it gets paid out precisely in the weeks a thin roster can least afford a zero.

Variance also stacks, and this is where drafters lose the thread. Six wide bands in a starting lineup do not average out across 17 games the way they average out across a thousand simulated seasons; you get one draw, and the swing on a lineup of six coin flips is enormous in both directions. Two or three wide rows in a starting nine is a posture; six is a lottery ticket with an entry fee.

Payout structure moves the answer more than most drafters admit. Winner-take-all pays the ninetieth percentile and nothing else, which makes width close to free. A league paying the top three pays medians and floors, and the same wide row that was free becomes expensive.

Band width also settles which tier breaks are real. Two players in one tier whose bands overlap across most of their width are the same buy at the same price, while a break separating two narrow, non-overlapping bands is a genuine cliff worth reaching over. That argument is made in full on the tier-break page.

The bottom of the board is where width is cheapest. A wide P90 at pick 140 costs nothing a roster would otherwise start, which is the entire economic case for the archetype; the sleeper screen is this same width argument run against late ADP.

Where This Sits In Your Draft

Band width is the second pass. Sort by price, cut the board to the players your picks can actually reach, and read the widths across whatever survives that cut, letting roster shape choose between the wide rows and the narrow ones. The P10 and P90 columns sit beside the median on the draft rankings board, and they move whenever the model is rerun.

One caution about the output itself. A wide band is the most comfortable thing a projection system can publish, because a band that covers everything is technically never wrong, and a model that answers every hard player by widening the band has bought itself insurance at the reader's expense.

Whether ours does that is a question the record answers. Hit rates inside the bands, the calls that landed outside them, and the seasons where a narrow band broke anyway are all posted on the public accuracy ledger — read the record, then decide how hard to lean on any of it, including this piece.