rookie projection model

how we project first-year NFL players before they have NFL data

rookie projection is the hardest problem in fantasy football modelling. you are predicting NFL performance for players who have never played in the NFL, using college data that doesn't map cleanly to pro production, for situations (landing-spots) that aren't known until draft weekend. every model that says "high confidence on this rookie" is lying.

ours says it too, sometimes. here's how we try not to.

the two-stage architecture

the model runs in two stages because the available information changes drastically between pre-draft and post-draft.

stage 1: draft capital prediction (pre-draft)

inputs: college production metrics, athleticism scores, consensus draft scout rankings, position

target: expected draft position (round, pick range)

why this matters: draft capital is a strong proxy for NFL team conviction. teams spend first-round picks on players they believe will start immediately. a WR drafted in round 1 will get target share; a WR drafted in round 5 probably won't. the stage 1 model translates college signals into the draft capital signal the NFL market has already priced.

college features used:
- production efficiency: yards per route run, receiving TD rate, drop rate
- usage rate: target share within college offense, snap percentage
- explosion scores: YAC rate, yards per reception, vertical burst at combine
- career trajectory: year-over-year production growth (breakout age signal)
- competition level: conference adjustment factor (see below)

stage 2: fantasy projection (post-draft, with landing spot)

inputs: stage 1 draft capital prediction + actual draft result + landing-spot context

target: projected fantasy points per game for season 1

landing-spot variables added:
- depth chart position (starter vs. backup vs. unclear)
- target share opportunity (based on team's historical distribution + departures)
- backfield competition (for RBs: RBBC vs. workhorse vs. committee)
- team passing volume (low-volume offenses suppress all WR/TE production)
- scheme type: pass-heavy, run-heavy, air-yards-centric vs. RAC-heavy

the post-draft model does not run until landing spot is known. a pre-draft projection without landing spot is stage 1 only, and it shows as lower confidence.

conference adjustment

not all college production is equal. a WR putting up 1,400 receiving yards in the SEC is a meaningfully different signal than the same number in a MAC offense. the conference factor is a multiplier applied to raw production metrics.

the factors are derived from the historical production translation rate: how often does a player from conference X at production level Y produce at position Z in the NFL?

key adjustments:
- power conferences (SEC, Big Ten, Big 12, ACC, Pac-12): multiplier near 1.0
- Group of 5 (AAC, MAC, Mountain West, Sun Belt, C-USA): discount of 0.82-0.91 depending on position and metric. skill-position production discounts more heavily than size/athletic scores.
- FCS: discount of 0.70-0.80. small samples; use with caution.

the conference adjustment is one of the highest-leverage features in the stage 1 model. ignoring it is how you end up drafting UTSA wide receivers in round 6 of your fantasy draft.

athleticism scores

raw combine metrics (40-yard dash, vertical, broad jump, short shuttle) are combined into position-specific composite athleticism scores. the composite is more predictive than any single metric.

for WRs: the composite weights 40 time (speed), vertical (explosion), and short shuttle (change-of-direction). broad jump adds less marginal value once speed is controlled for.

for RBs: 40 time matters less; burst score (10-yard split) and contact balance (self-reported, supplemented by college film grade) matter more.

for QBs: combine athleticism scores have low predictive power for fantasy outcomes. this reflects the reality that what makes a QB productive in fantasy (arm talent, decision speed, touch) doesn't show up in a 40.

confidence on rookies

almost every rookie projection starts with a confidence score below 0.70. this is correct. the reasons:

  1. no NFL sample: confidence is partially driven by games in the signal window. zero NFL games = no window.
  2. high variance in college-to-pro translation: even the best college WRs fail at a high rate. the distribution of outcomes for a first-round WR in his first NFL season is genuinely wide.
  3. landing-spot uncertainty: even post-draft, situations change. training camp, preseason injuries, scheme adjustments — all of these shift the projection before week 1.

we consider a rookie projection above 0.65 confidence meaningful. below that, treat it as a range estimate and round-value accordingly.

what the model consistently gets right and wrong

tends to get right:
- identifying which rookies are in high-opportunity situations (target share, backfield role)
- penalizing players who land in low-volume offenses
- flagging the "drafted high, bad situation" case (QB drafted round 1 but sitting behind a starter for 2 years has very different fantasy value from a starter from day 1)

tends to get wrong:
- year-1 breakout candidates in unusual schemes. the model learned from historical distributions; truly novel scheme fits produce upside the model can't see
- receivers with elite separation who go to run-heavy teams and then see the scheme shift mid-season
- QBs: the position is hard. don't draft rookies QBs in redraft leagues based on our model output. this is not the model talking. this is experience.

frequently asked questions

why don't you have confidence above 0.85 on any rookie?

because that would mean the model has high certainty about a player with no NFL data. no honest model has that. if another rankings service gives a rookie a top-10 WR rank with high confidence, ask them to show their calibration methodology.

Harold Fannin Jr. is ranked TE4. isn't that crazy?

the model is reading strong usage signal in the CLE system combined with a full-season projection and a young age multiplier. the confidence score (0.97 on that specific projection) is high because the feature coverage is complete. "crazy" is relative to how the market is pricing him. if ADP puts him at TE12 and the model says TE4, that's a market inefficiency worth examining.

can I see draft capital predictions for undrafted free agents?

UDFAs are included in the rankings through a separate pathway that uses FantasyPros ECR as a landing-spot proxy. their confidence scores are very low (0.30-0.45). treat them as speculative.