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. this page covers the approach; the specific features and calibration stay proprietary.

why two stages

the available information changes drastically between pre-draft and post-draft, so the model works in two stages.

before the draft, it reads college signal and athletic profile to estimate where a player will be drafted. draft capital matters because it's a strong proxy for NFL team conviction: teams spend premium picks on players they intend to use, and usage is what produces fantasy points. a player projected to go early is priced very differently from one projected to slide.

after the draft, the actual draft result and the landing spot get folded in: where the player sits on the depth chart, how much opportunity the offense has to give, and what kind of offense it is. a talented player in a crowded, low-volume situation projects lower than the same player handed a role. the post-draft projection doesn't finalize until landing spot is known; before that, a rookie shows as lower confidence.

not all college production is weighted equally, and athletic testing is folded in where it's predictive and ignored where it isn't. the details of how are part of the model, not this page.

confidence on rookies

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

  1. no NFL sample. confidence is partly driven by usable history, and a rookie has none.
  2. wide outcomes. even elite college players fail at a meaningful rate; the distribution of first-year results is genuinely broad.
  3. landing-spot uncertainty. even after the draft, camp battles, preseason injuries, and scheme changes move the projection before week 1.

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

what the model tends to get right and wrong

tends to get right:

  • identifying which rookies land in high-opportunity situations (role, volume, target share)
  • penalizing players who land in low-volume offenses
  • separating "drafted high, bad situation" from "drafted high, plug-and-play role"

tends to get wrong:

  • year-one breakouts in unusual schemes the historical data hasn't seen
  • talented players whose situations shift mid-season
  • rookie quarterbacks. the position is hard to project in year one; don't draft rookie QBs in redraft leagues on the strength of a model number. that's experience talking, not the model.

frequently asked questions

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

because that would mean the model is certain about a player with no NFL data, and no honest model is. if another service gives a rookie a top-ranked grade with high confidence, ask to see their calibration.

the model has a rookie ranked way above ADP. is that a mistake?

not necessarily. when the model reads a strong situation and a full-season role, it can land well ahead of where the market is pricing a name. that gap is the point: it's a candidate market inefficiency worth examining, weighted by the confidence score attached to it.

can I see undrafted free agents?

UDFAs are included through a separate pathway and carry very low confidence scores. treat them as speculative.