Methodology

Nine seasons replayed at ten teams and full PPR, ranked the way the board ranks players: drafted players finished closer to this board’s order than to the draft position of free mock drafts in seven of the nine seasons, and closer than paid expert consensus in all five seasons it publishes for. The edge is widest after pick 60. Inside the first 36 picks the board is level with the market. In dynasty, tested against three seasons it had never seen, the top twelve priced within 3% of what they returned in a 1QB league and within 10% in superflex.

That record belongs to the full model, published each night. Below is each part of it: what it is, what it’s built from, what it beat and what it can’t see. Then what was tried and dropped, and the known limits.

Value, everywhere on the site, is points above a replacement starter at your league’s size, slots and scoring. Dynasty value is the same number added up over five years and discounted for how far away each year is. The market’s number sits beside BX’s and never feeds into it.

The model

A season projection is one chain: how much the offense will do, how much of it is his, what he does with it, how many games he plays, and what that is worth in your league. The five parts below are those five steps, and the changelog files every change under the step it came from.

Team

What it is: How much each offense will do this season: pass attempts, carries and targets, one pool per team per stat. Every player’s volume is a share of his team’s pool, so the team number is the ceiling on everyone in it. Beside it, a rating for each team’s offense and defense, adjusted for who it faced, and a rating solved from the betting market.
Built from: The public weekly stat lines, the schedule, and the posted spread and total for every game. The unit ratings on the NFL pages are solved from the same weekly lines with each unit adjusted for its opponents, then blended with where the team finished last season, counted as if it were four games, so a rating after one game is one part this season to four parts last. The same page shows a rating on this season alone beside it. The market ratings are solved from every spread posted this season.
How it compares: The record at the top is earned on playing time and roles, which Team, Role and Games carry between them. The check on team scoring is on the schedule: across 2,639 games from 2016 through 2025, actual scoring ran 0.4 points a game above the posted total, and the split between the two teams showed no lean.
What it cannot see: Why a pool moved. There is no news feed, so the changelog can say Denver’s target pool fell and who took the loss, and never why. Preseason games are in no source the model reads. Kickers and defenses carry a team projection shrunk hard toward the middle, because the spread anyone can predict before the season is about a point a game.

Role

What it is: His share of his team’s pool: of the carries, of the targets, of the pass attempts. Role and team are kept apart so a change in one is never reported as the other. In one August publish Atlanta’s carry pool fell 8.9% and Bijan Robinson absorbed 36 of the 42 carries lost, which is a different fact from Atlanta giving his carries to someone else.
Built from: Depth charts, the injury report and roster status, and the player’s past seasons of usage. A rookie’s role starts from where he was drafted; the rookie models move him off that, and Rookie picks below says how far to trust the move.
How it compares: On per-game scoring alone the market projects running backs, receivers and tight ends slightly better than BX does, so the record at the top is earned on playing time and roles, and mostly after pick 60. In the weekly tool a share updater that reads the season so far closes 39% to 74% of the gap to the expert consensus at running back, receiver and tight end from week five on.
What it cannot see: Snap counts, routes run and red-zone share are not inputs. A usage number joins the model only after a test on the nine seasons says it helps, and that test has not been run for any of those three. A depth chart is read as published; a coach’s stated plan is not read at all.

Efficiency

What it is: What he does with the volume: completions and yards per attempt, yards per carry, catch rate and yards per target, touchdown rates, interceptions and fumbles. What is published is the counting line, and your league’s scoring turns it into points when the page loads, so a six-point passing touchdown or a tight-end premium prices the same line differently and nothing is re-projected.
Built from: His past stat lines and his position’s rates.
How it compares: Nothing: on per-game scoring at running back, receiver and tight end the market is slightly ahead of BX. The board’s record comes from team volume, roles and games.
What it cannot see: Yards after the catch over expected and rushing yards over expected are built from play-by-play, and the model reads no play-by-play. Anything about how he produced the line that the line itself does not carry.

Games

What it is: How many games he plays. Every projection is per game times games, and value is per game over replacement times games, so a player who is brilliant for nine games did not win you the other eight weeks and the number says so. A player projected for fewer than eight games orders on season points rather than per-game value, so a backup’s mop-up rate cannot jump a starter. A player out for the season is off the draft board.
Built from: The injury report (Out, Doubtful, Questionable) and roster status (reserve, physically unable to perform, suspended) as two separate facts, never mapped onto each other, since a suspension is not an injury. In season, the probability he plays is fitted from the Sunday morning report.
How it compares: The fitted probability that a player suits up beats the rule of thumb. The weekly line you see is his if-he-plays line times that probability, so a doubtful player carries a fraction of it and a player ruled out carries none.
What it cannot see: An injury that has not happened. The games number is an expectation, so even a healthy starter is projected for fewer than seventeen and a season-ending injury in week three is in nobody’s number. Nothing after the Sunday morning report reaches the weekly line.

Value

What it is: Points above a replacement starter at your league’s settings. Replacement is the best player at his position who does not start in your league, found by walking your starting lineup down the board, flex and superflex included. At ten teams with one flex that is the eleventh quarterback, the twenty-second running back, the thirtieth receiver and the eleventh tight end. A gap of eight points between neighbors opens a new tier. The arithmetic, on real players at three league shapes, is on What Value means.
Built from: Your league: teams, starting slots, scoring. Nothing else. One function prices the board, the trade calculator, the mock room and the waiver wire, so a player’s value matches across all of them.
How it compares: The formula it replaced: teams times starters, plus a fixed share of the flex and 80% of a superflex seat to a quarterback. Both fixed numbers were measured wrong. A superflex seat goes to a quarterback nearly every time, and the flex share runs from nearly all receivers in a shallow one-flex league to almost a third tight ends in a deep one.
What it cannot see: The three-season dynasty test below was run against the earlier formula, and this way of finding the replacement player has not been tested on its own. The past seasons on file have no sacks-taken column, so a board for a league that scores sacks has never been tested. In dynasty the replacement player is built into the published value and has no name to show.

The tools

Five things built on the model. Each reads the same published numbers and the same replacement rule; none of them projects anything of its own.

Weekly projections

What it is: Every player projected for the week, and for every remaining week, restated every day. The Sunday morning restate is the last one before kickoff. The number at your scoring is what he does if he plays. The chance he plays is published beside it, never multiplied into it.
Built from: The season model, the depth chart, the injury report, and what has happened so far this season.
How it compares: Replayed over nine past seasons, it beats last week’s line, the season average, and the season projection on its own. The expert consensus is still ahead, by about one to three right start-or-sit calls per hundred at running back, receiver and tight end from week five on; before games played can be counted, the gap is wider. Against expectation holds the projection that stood before each game against what happened, pinned to each player’s own kickoff, so nothing published after a game counts.
What it cannot see: Anything after the Sunday morning report, and anything that is not in a depth chart, an injury report or a stat line.

Dynasty values

What it is: A player’s value over the years ahead: the season value carried forward year by year on how players of his age and position hold their value, each year discounted, and added up over five years. Three discount rates are published: win now at 0.65 a year, balanced at 0.80, rebuilding at 0.92. The dial is a preference about now against later.
Built from: The season projection as the seed. His draft slot for a share of the first year, which is why a mid-career back’s first year is partly where he was drafted. His age, and the measured curve of how each position ages, with running backs and receivers 28 and older on their own group’s curve.
How it compares: Tested against three seasons it had never seen: the top twelve priced within 3% of what they returned in a 1QB league, within 10% in superflex. Running backs, receivers and tight ends order at least as well as the startup draft market, and the board was ahead of the expert-consensus dynasty ladder in the first year out in all three seasons.
What it cannot see: Elite quarterbacks returned more than BX prices them at, and superflex quarterbacks sit below what the trade market pays. Players past thirty sort better on the market’s age discount than on BX’s. Later-year values for mid-board players run high at quarterback, tight end and receiver; the top of the board is unaffected.

The market beside BX

What it is: The market’s number beside BX’s on every surface, and the gap between them as a verdict. Where the market takes him later than BX would, he is cheap. Where it takes him earlier, let someone else pay. The market never enters the BX number, so the two can disagree.
Built from: Draft position from free mock drafts in several redraft formats, KeepTradeCut’s dynasty prices in 1QB and superflex, and the paid expert consensus. A league with no matching market, tight-end premium for one, shows the nearest feed and says so.
How it compares: The record at the top is against these markets, and the gap is widest after pick 60. On rookies, where the BX order disagrees with the draft order: at running back the disagreement beats what a lucky reshuffle of draft position produces, and did so in every class the model had never seen. At receiver it pointed the right way in every such class by margins a reshuffle could also reach. At quarterback and tight end it cannot be told from noise and reads as commentary.
What it cannot see: Whether a gap is exploitable. A price above what a pick’s slot returned is a reason to look, not proof of a bargain. The one market fact with a clock on it: as of mid-August 2026, 2027 first-round picks sell for 1.16 to 1.27 times the matching 2026 pick, and six years of daily pick prices never left 0.87 to 1.13.

Rookie picks

What it is: What a rookie draft slot is worth: ten years of real rookie classes, what the players taken in each slot range returned, in points above replacement over three years at the balanced discount. The first three picks returned nearly three times the next three and their ranges do not overlap. Below the top, ten classes cannot tell some neighboring slots apart, so picks are quoted in shared rungs and constant within a rung. A future pick is priced at its round’s average slot and discounted at the balanced rate for each year away.
Built from: Ten rookie classes, the NFL draft, and the dynasty values of every player in them. A rookie’s own value carries no market price: it comes from the rookie models and the pick curve alone, and the Evidence tab on the rookie board says how far to lean on each position’s flag.
How it compares: The curve is a record of what past picks returned, not a forecast, so there is nothing to test it against. It does answer one question: round one is a tier only at the top, since the first three picks and the next three do not overlap. The market’s asking price sits beside the measured return, and at most bands the market asks less than history returned.
What it cannot see: Which slot a future pick lands in, and how strong a class is before the draft: class strength turned out to be a fact about where draft capital went, and cannot be forecast before it is spent. A pick’s BX value is the ten-class average and says nothing about how strong that one class is. The curve stops at slot 36, so a fourth-round pick in a twelve-team league shows a dash. The first overall pick alone has returned everything from a bust to six times a typical season.

Your league

What it is: Your league as the frame: every roster priced at your settings, who holds each player, the waiver wire with this week’s number, trades priced in the same value as the board, and a mock draft room whose opponents draft with the pick-to-pick randomness of real drafts.
Built from: A connected Sleeper, ESPN or RTSports league. For the mock draft room, pick-to-pick randomness measured on real MyFantasyLeague drafts.
How it compares: The odds a player survives to your next turn are tuned against picks that happened; the untuned version ran about 15% too hot. The cost-of-waiting number was checked against real outcomes across 1,120 draft turns. The mock room’s advice was tested against drafting in the market’s order over nine seasons and hundreds of drafts a season, counted in championships: a tie, though the test could only catch a large difference. What the mock room adds is warning: who will be gone before your next pick, and what waiting costs.
What it cannot see: Humans who adapt, trades and in-season pickups: the test used simulated opponents who do none of those. What the other manager will accept.

Tried and dropped

Ideas that were tested and dropped, or kept out on purpose. Each stays out unless a new test on past seasons shows it helps.

Market prices as an input. Draft position and trade values sit beside the BX number and never inside it, because a model fed the market cannot disagree with it.
Public prospect models and scouting grades as inputs. They are checks against the BX rookie order, and a disagreement is a question to investigate rather than an error to correct.
Play-by-play. The model reads weekly and season stat lines only, so yards after the catch over expected and rushing yards over expected are not inputs.
Rookie values before the NFL draft. Class strength turned out to be a fact about where draft capital went, so the pre-draft lane was dropped.
Adjusting a future pick for how its team looks today. That is a team-strength forecast a season ahead, which has never been built.
The market’s discount on future picks. BX values a pick on what that slot has returned, not on what picks sell for.
The win-now and rebuilding dial as a trade rule. As an accept-or-reject rule it made no difference over the replay, so it is a matter of taste, not an advantage.
The old replacement formula: 80% of a superflex seat to a quarterback and a fixed flex share. Both numbers were measured wrong.
A flat top-two-per-position rule for which backups count in a weekly lineup value. A committee’s third back does nearly twice the work of a bell-cow’s second, so the rule follows the backfield instead.
Significance labels on rookie flags. With only ten classes they showed up on pure noise, so each flag carries a strength rating in words instead.
Quoting picks at three years against players over a full career, and stretching the pick curve past its three measured years with a model ratio. The first understates every pick and the second invents evidence, so when a pick joins a trade both sides are quoted at three years.
The odds a player finishes first at his position. BX overstates them and the correction did not fix it, so they are not shown.
Snapping a connected dynasty league to the nearest published shape without saying so. The pairing is printed beside the numbers.
Players out for the season on the draft board.

Known limits

Draft advice in real leagues, with humans who adapt, trades and in-season pickups. The test above used simulated opponents.
Whether the ask-versus-returned gaps on rookie picks are exploitable.
Whether these values make anyone money in a real dynasty trade. That depends on the other manager, whom no test reaches.
This way of finding the replacement player, on its own. The three-season dynasty test was run against the formula it replaced.
Snap counts, routes run, red-zone share and year-over-year direction as inputs.
A league that scores sacks taken.