CombineMethodology
How a player gets a number
No part of the ordering is hand-placed. A player’s rank is the output of a chain that starts with expected volume and ends with the value they hold above the player your league could get for free. Every step is inspectable, and the model is allowed to be wrong in public.
CombineStep one
Volume becomes points
Expected usage is converted into fantasy points under your exact scoring rules. This is the only place scoring format enters the model, and everything downstream recomputes from it.
Worked example — Brock Bowers
- Expected games
- 16
- Targets
- 148
- Projected points
- 273
- Above replacement
- +127
In full PPR he ranks 5 overall. Remove receptions from the scoring and the same usage produces a different answer: 13 overall. Nothing about the player changed — the question did.
CombineStep two
Replacement level, not raw points
A quarterback who scores 350 points is not more valuable than a running back who scores 280, because your league can find another quarterback and cannot find another back. Combine ranks by value above the last player who would actually start.
Starting demand
Teams multiplied by starting slots at each position, with flex and superflex allocated greedily to whichever position's next player scores highest. That is what makes superflex genuinely change quarterback priority instead of applying a fudge factor.
The baseline
The three players straddling the last startable slot are averaged, so one outlier at the boundary cannot swing an entire position's value.
Value above it
Projected points minus that baseline. It is the only number that answers 'what does drafting this player actually buy me over doing nothing'.
CombineStep three
One evaluation, many inputs
Combine turns role, projection, risk, replacement level, and league settings into one readable decision surface. The number stays explainable because every major input remains inspectable.
What feeds the model’s estimate
- · Expected volume and role security
- · Team scoring environment, pace, line and quarterback play
- · Efficiency the model expects to persist
- · Durability and expected games
- · Positional replacement level under your settings
What shapes the recommendation
- · League scoring and roster configuration
- · Positional scarcity and replacement level
- · Risk, durability, and role security
- · Expected availability at the next decision
- · Your roster's remaining needs
The recommendation is a starting position, not an oracle. The reasoning stays visible so you can apply your own league knowledge.
CombineThe components
What the Combine Score is made of
The headline score is a weighted blend of league-adjusted value over replacement and six scouting components. It is a summary, not an oracle, and every subscore is shown separately on the player profile.
| Component | Weight | What it measures |
|---|---|---|
| Production | 24% | Prior fantasy production, adjusted for the role it was earned in. A back who scored on 320 touches and a back who scored on 180 are not the same finding. |
| Opportunity | 30% | Expected volume: targets, carries, route participation, snap share, red-zone and goal-line work. The heaviest single component, because volume is the most predictable input in fantasy football. |
| Environment | 14% | Offense quality, pace, offensive line, quarterback play, and implied team scoring. Volume in a bad offense is worth measurably less. |
| Efficiency | 12% | The part of last season's efficiency the model expects to persist, which is much less than the part that showed up. Touchdown rate regresses hardest. |
| Role security | 12% | How safe the role is from committee, competition, scheme change or benching. This is what separates two players with identical projections. |
| Durability | 8% | Games-played expectation from injury history and workload. Applied to expected games rather than to per-game output. |
Risk and confidence are modeled separately and never folded into the score. Risk answers “how likely is this to go wrong”; confidence answers “how much should you trust this estimate at all”. They move independently — a healthy rookie with a locked role is low risk and low confidence at the same time.
CombineUncertainty
Where the model refuses to pretend
Fake precision is the most common failure in this category. A ranking that puts one player 0.03 points above another is asserting a resolution nobody has.
Tiers, not places
Tiers are found from real breaks in the value distribution. Inside a tier, draft cost should decide the order, not the ranking.
Ranges, not points
Every projection carries a floor and a ceiling derived from the player's own upside, floor and durability grades.
Confidence bands
Confidence falls for rookies, unsettled roles, team changes, coaching changes, and when sources disagree.
Named assumptions
Where the model changes its recommendation it says which component drove it and by how much, so you can reject the reasoning rather than the number.
CombineLimits
What this model does not do
- It does not predict the future. It produces a defensible prior and shows you what it rests on.
- It does not model weather, game script, or in-game injury. Those are outside what a preseason valuation can honestly claim.
- It does not know your league-mates. If the manager at pick nine always takes his own team's players, that is information the model cannot see and you can.
- Kickers and defenses remain in the rankable player pool when the active board has usable evidence for them, but their projection side is intentionally conservative because the current model does not pretend to have a live week-specific special-teams forecast.
- Market-ranking evidence can refresh separately from the football projection/status snapshot. A newer market rank never implies that projections, injuries, news, or schedules were refreshed at the same time.
Demonstration build · 2026 preseason · fixed reference timestamp
CombineQuestions