What this is, and what it is not

BTB Data projects opportunity: targets and carries. It does not sell picks and does not claim an edge on game outcomes. That is a measured position. The one fully backtested NFL game model in our stack graded 49.96% against the spread over 1,311 graded games from 2021 to 2025, with no confidence bucket clearing break-even. The market prices game results well. It prices where the ball goes less carefully, and that is what a fantasy lineup and most player props turn on.

How a projection is built

1. Team volume. Each team's pace and neutral pass rate, shrunk toward the league mean by how many games it rests on, then tilted by the spread. A favourite runs more; a big underdog throws more.
2. Each player's share. A fitted share model. Every one of a team's targets, and separately every carry, goes to one of its active players in proportion to a score, and the score's weights were fitted on 2024 and 2025: last season's share, which fades as this season's games add up; his target and carry share over his last three and five games; his route share last season (route data is published only after a season ends); his snap share and whether it is rising or falling; his place in the team's pecking order at his position; whether he has changed teams; and, for a quarterback's carries, whether he is the starter.
3. How the share model scored. Fitted on 2024 and scored on 2025, it missed by 3.52% less than the old hand-built split on carries and 2.04% less on targets, clearing the 2% bar we set on both; we serve it on both. The replay on the record grades it on each week's active roster and injury report; for 2024 and 2025 that replay is in-sample for its weights, and 2026 is not.
4. Shares add up. The split is within the team, so a team's shares sum to one by construction. Without that, three backs on one roster can each project for twenty carries.
5. Who plays at all. Out, Doubtful, IR and off-roster players leave the split, and their share goes to teammates in proportion to their scores. A Questionable tag lowers a player's score by as much as questionable players' usage fell in 2024-25. The game-day inactive list does not reach us yet: a healthy scratch, or a Questionable player ruled inactive 90 minutes before kickoff, keeps his projection, and his teammates read a little low.
6. Yards and touchdowns. Yards per carry and per target come from each player's role, not his results: the expected yards his opportunities have carried over three seasons (how deep, where, what down and field position), shrunk toward his position. That beat the flat league rate for receiving and rushing yards in both seasons replayed (2024 and 2025). His own yards per touch, over three seasons and shrunk toward his position, beat the flat rate too for receiving yards in both seasons and for rushing yards in 2024, but lost to it on rushing yards in 2025, and his role beat his own rate in three of the four. Catch rate keeps a small pull toward his own season. Receiving yards are then scaled to the quarterback's projected passing (below). Touchdowns are his share of the team's projected opportunities times the team's expected touchdowns. Red-zone share was tested as the basis: season-to-date, it forecast touchdowns worse than plain touch share in both 2024 and 2025; shrunk toward his touch share, it was a hair worse in 2024 and a hair better in 2025, not better in both, so touch share stays and red-zone share is shown for context only.

Quarterbacks

A quarterback's line starts where the play does, in three matchups. Volume: his team's dropbacks against what this defense concedes, blended with his own dropbacks per start (a scrambler turns dropbacks into runs), tilted by the spread. Protection against the pass rush: the sack rate he and his line give up against the rate this rush gets home, combined on the odds scale. A sack is a dropback that never becomes a throw. The passer against the coverage: completion rate, yards per attempt, touchdown and interception rates, each his own shrunk toward a replacement-level passer by how much history he has, against what this secondary allows.

Two things the data would not support, and so the engine does not do: a separate penalty on yards per attempt for pressure, which helped one season and hurt the other, and the touchdown rate a defense allows, which carried nothing. Receivers' yards and catches are then scaled to the quarterback's passing, which beat pricing every target at the league rate at every position in both seasons replayed. In that sum a receiver with a yards line counts at his mean, his line times a fixed ratio for his position: a line is a midpoint and midpoints do not add, so the receiving yards shown do not add up to the quarterback's exactly. The evidence, piece by piece, is on the record.

The market

The prop board is a multi-book capture of each line. Its probability is a cross-book median, de-vigged where enough books post both the over and the under; where only the overs are posted (62 of this week's 144 lines), it is the middle of the over prices with the vig still in, which overstates the over, and the site marks it "vig in" rather than calling it fair. Where the books post a consensus line on a number we project, the site serves the line itself, with our engine's number beside it and no gap (this week: receptions, receiving yards, rush attempts, rushing yards, anytime touchdown, passing attempts and passing yards). Where we project the same quantity as a line and do not serve the line (this week: passing completions), the site shows both numbers and the gap between them. The gap is measured from our median, because a book sets its line at its median; a mean against a half-point line on a small count (interceptions, passing touchdowns) sits above it every time, so those get no gap. A game that has kicked off shows only what was logged before kickoff. It never turns our projection into an expected value. A fair line generated from the same projection you then grade against is positive by construction; the suite has shipped that mistake before and removed it.

The week

The board rebuilds through the week as practice reports and lines move, and again after each inactive list is posted: Sunday 10:55, 11:20, 14:15, 14:40 and 18:15 Central, and Monday and Thursday 18:15. Those builds carry the latest injury report; the inactive list itself is not in our player feed yet. Every projection is written to a ledger with the time it was logged. The ledger refuses rows for games that have kicked off, and only rows logged before kickoff count toward the record. The record grades the player-weeks where he took the field (a box-score row or an offensive snap): a player who sat is left out, because his easy zero would flatter the average miss, and so is one who left the game injured (an injury in the play-by-play and his offensive snap share under 60% of his average over his last three games, or under 50% when he has no earlier game with a share), because we projected a game he did not get to play. Both are fixed rules read from the game's own records on the Tuesday after, never a choice about a single player, and they apply whether we were close or far. On the record, "better by" is the baseline's miss minus ours on the paired rows only: the ones where the player had an earlier settled week to average.

Each game page shows the forecast for kickoff: temperature, wind and gusts, and the chance of rain. US games come from the National Weather Service; games abroad from MET Norway (weather data from MET Norway, CC BY 4.0). A fixed roof reads "Indoors"; a retractable roof shows the forecast and says the roof may be shut. Once a game kicks off, the page keeps the last forecast taken before it. The weather is shown as context only: no projection, price or range on the site uses it.

The record and the tiers

Projections carry no grade chip (since September 28, 2026). The grades live on the record, and every one there (tested, proven or ungraded) is computed from the ledger by the code that builds the site; no page may call a number graded unless that code says so. PROVEN needs every position to clear the bar on its own, over several weeks, and to beat each player's own prior average on the same rows. The full rules and every graded number are on the record.

Specialty stats

The Role group on a player's profile, and the accuracy numbers on a quarterback's, say when and how he is used and how well he throws. Each covers the season his profile shows (this one so far, or last season in full before he has played this year), ranked among qualified players at his position.

First-read target share

His targets on throws charted as the QB's first read, over all of his team's first-read targets in his games.

Designed-target share

The same, for throws charted as designed (screens and quick game).

3rd-down / passing-down target share

His targets on 3rd down (or on 2nd & 8+ and 3rd/4th & 5+) over his team's.

Deep / red-zone / end-zone target share

20+ air yards; snaps at the opponent's 20 or closer; throws whose air yards reach the goal line.

Checkdown share of targets

His targets charted as the QB's checkdown, over his targets with a read charted.

Contested catch rate

His catches on targets charted contested, over those contested targets.

Created receptions

The share of his catches FTN charted as created.

YAC / reception

His yards after the catch over his receptions.

Dominator rating

The average of his share of the team's receiving yards and receiving touchdowns. For a back, the same over scrimmage yards and touchdowns.

True yards per carry

Every carry counted at no more than 10 yards, so one long run does not carry the average.

Yards / touch

His scrimmage yards over his carries plus catches.

Game script

His team's average score margin at the snaps where he got the ball. Positive means ahead.

Catchable-throw rate (QB)

Aimed throws charted catchable, over aimed throws (throwaways excluded).

Bad-throw rate

Pro Football Reference's bad throws over aimed throws.

Interception-worthy rate

FTN-charted interception-worthy throws over charted attempts.

Completion %

Completions over pass attempts.

True completion %

Completions plus drops over aimed throws, so a drop counts as a completion the quarterback earned.

Pressured rate

Pro Football Reference's pressures over his dropbacks.

Throws to first read / checkdown / off-script

The share of his aimed throws charted as each.

Time to throw, tight-window throws

Next Gen Stats.

Receiver separation on his throws

The average Next Gen Stats separation of the receivers he targeted, weighted by his targets to each.

Charting is FTN Data's, via nflverse (CC-BY-SA 4.0); bad throws and pressures are Pro Football Reference's; time to throw, tight-window throws and separation are the NFL's Next Gen Stats.

The role card

The role card on a player's page is context for your own research, not a prediction. It shows how a player has been used, how his offense spreads its targets, what this week's defense has given up per target to players of his kind, and where those signals disagree. None of it feeds a projection, and none of it has been checked against results. The Matchup score on the same page is a separate number and is not a prediction either.

His window. Every number is from his recent games with his current team, never from this week's game, so a game that has started shows the card it had before kickoff. Shares are his count over his team's in the same games.

Roles name how he is used:

Alpha

Gets the ball first, and it isn't close.

1A / 1B

Shares the top of the target tree.

Lead option

Tops a passing game that spreads it around.

No. 2 option

Steady work behind the lead.

Rotational

Mixes in.

Hub tight end

The passing game runs through him.

Seam stretcher

A receiver who lines up at tight end.

Red-zone tight end

Quiet between the 20s, busy near the goal line.

In-line blocker

Plays a lot, blocks first, catches short.

Depth tight end

Situational snaps.

Bell cow

Takes most of the carries.

Lead back

The top of a shared backfield.

Goal-line hammer

Gets the carries inside the 5.

Third-down back

Earns his work catching passes.

Committee back

One piece of a split backfield.

Change of pace

A few touches a game.

Role still forming

Fewer than three games with this team, or fewer than 40 team targets in his window.

Whether a tight end lined up in the slot or next to the tackle is not in free play-by-play, so tight-end types are an approximation.

Personnel and route participation come from who was on the field, which is published only after a season ends. In season they are last year's numbers, and the card says through which week.

The defense. What this week's defense has given up per target to throws and receivers like his. A defense is called soft or stout only on a solid sample and a clear gap from the league.

Where the signals disagree flags spots where the defense's tendencies and his usage point different ways. Each flag names both inputs and their samples. They are prompts, never checked against results, and never a pick.

BTB Trust on props

The Props board ranks every line by BTB Trust: how much a player's recent record against that exact line backs it, scored 0 to 100 within the market. It rates the line, not the player, so 100 is the line on that market this week his record backs most.

A market the score cannot grade (longest plays, first touchdown scorer, quarter and half lines, kickers, defense), or a player with too few games, gets no score: a dash, never a filler. Like the rest of the board it is descriptive, not a forecast or a pick: a player's record against a line is not shown to beat the book's price.

BTB Range

Every pregame player gets a floor and a ceiling: his 10th and 90th percentile outcomes for targets, receptions, carries, receiving and rushing yards, and full-PPR fantasy points. They start from the spread we fit for his position, the same one behind the price ladders, around his projected number. Then his own last 16 games, across seasons, widen or narrow it: a player whose games have strayed further from his own average than the position's do gets a wider range, pulled back toward the position's on a short record. Boom and Bust are the chances of 25+ and under 12 points for a quarterback, 20+ and under 6 for a back or receiver, 15+ and under 5 for a tight end. On the Fantasy board, DK 5× and DK <2× are the chances of five times and under two times his DraftKings salary in points, without DraftKings bonuses. BTB Consistency ranks how tight his range is against every player at his position this week, 100 the steadiest; on the Props board it shows as Swing (low, mid or high). Consistency is relative to his own median, so a high-volume player can rate steady while his raw totals still move a lot; the floor and ceiling show that raw swing.

Before any result was looked at, we wrote down how it would be checked: fit on 2024, check on every 2025 game, and call a market settled only if the player's own spread beat the position's alone and its 80% range held 75-85% of the time. Targets and rush attempts cleared both; fantasy points, checked on calibration alone, held. Receptions and rushing yards did not beat the position's spread, so they show with an "early read" label. Receiving yards did worse with each player's own spread than with the position's, so its range uses the position's spread and is labelled "position range". Of four game-context ideas (the total, the spread, the opponent's pass rush, its tackling), only the game total cleared the bar, only for rush attempts, and only just: with 24 such checks, one could clear by chance. The range describes how wide his week can go. It does not change any price on the site, and it is not a pick.

Freshness

The site is static and rebuilds when the data does. The footer of every page gives the date of the latest refresh, and the table below gives the day each input behind this week was captured, because a dead upstream otherwise looks exactly like a quiet one.