Every fantasy football site you read pulls from roughly the same inputs: last year's stats, depth charts, and a smattering of training-camp buzz. That is why consensus rankings feel so similar across ESPN, Yahoo, and your favorite podcast. The names cluster, the tiers cluster, and the average draft position chart for any given week looks like a copy-paste job. Our edge is not that we have secret information. It is that we feed a wider feature set into a projection model — built on the same Tinker infrastructure we use for our betting models — and let the math weight things consensus underweights.
Why consensus clusters
Consensus rankings are an average of human guesses. Humans anchor. Once a player is "the WR8" in early summer, every subsequent ranker is reluctant to move him more than a tier without a clear narrative reason — an injury, a trade, a coaching change. The result is that rankings drift slowly, even when the underlying signal has shifted considerably.
Our Sharksnip projection model does not anchor. It rebuilds every player's projection from scratch each week using the same player_feature_store we power our prop models with: usage rates, route participation, target quality, red-zone share, snap counts, and team-level pace. When those inputs say a player should be a tier higher than consensus, we move them, with no loyalty to last week's number.
The three places we disagree most
Across our backtests on 2019–2024 seasons, the model disagrees with consensus most often in three repeatable spots. These are not "secret picks" — they are structural blind spots in how humans build rankings.
1. Aging veterans on bad teams
Consensus loves a name. A 31-year-old running back who hit RB1 numbers two years ago will keep getting drafted in the RB2 range long after his usage has quietly slipped. Our model weights recent route share and snap rate heavily and is brutal about fading veterans whose role has eroded, even when the headline still reads "starting RB on a real NFL team." Historically, RBs over 29 with declining snap share have hit RB2 finishes only about 22% of the time, despite an ADP that priced in 40%+ odds.
2. Year-2 wide receivers with target-share growth
The flip side is the wideout who graduated from a 14% target share as a rookie to a 22% share late in year one. Consensus tends to slot him in the WR4 range. The model treats target-share trajectory as a leading indicator and routinely projects these players a full tier higher. WRs with 25%+ target share retain that share year-over-year roughly 70% of the time, which is the kind of stability fantasy drafters consistently underprice.
3. New coaching staffs
Whenever an OC or HC changes, consensus assumes the old usage continues until proven otherwise. The model layers in coordinator-level pace, neutral pass rate, and personnel tendencies from the new staff's prior stop. Sometimes that drops a "consensus RB1" by an entire tier; sometimes it elevates an unheralded slot WR into the WR3 range overnight. We track those splits on the fantasy rankings page so you can see exactly where the model and consensus diverge.
How to read a model-vs-consensus delta
A delta is not a recommendation by itself. It is a flag to investigate. We grade deltas in three buckets:
- Half-tier (5–10 spot) gaps — Noise. Probably reflects ranker disagreement, not a real signal. Ignore in drafts.
- Full-tier (10–20 spot) gaps — Worth a second look. Usually one feature is doing the work; the model summary on each player tells you which.
- Two-tier (20+ spot) gaps — Either the model is wrong or consensus is. We display the top three contributing features so you can decide which side you trust.
The mistake to avoid is treating a 25-spot delta as automatically "draft this guy." Sometimes the model is leaning on usage data from a small sample, and consensus correctly sees the bigger picture. The point of surfacing the delta is to force the conversation, not end it.
Why this matters more than picks
Fantasy is a market. Beating the league means beating the room, and the room is using consensus rankings. Every draft pick where you take a player two rounds before consensus and they hit, you have captured a free win. Every pick where you let consensus push you off a player the model loves, you have donated equity to the rest of the league.
The same logic applies to weekly start-sit. Our start-sit tool shows the model's projection alongside consensus for every player on your roster. When the gap is large, the tool surfaces the reason — usage, matchup, schedule — so you can make a confident call instead of guessing.
What the model is bad at
This would not be an honest post if we skipped this section. The model is weakest in three spots:
- Rookies before Week 4. Limited NFL sample, and college-to-pro translation is noisy.
- Players returning from major injury. Usage features lag the eye test, so we tend to be slow to mark someone "back."
- Weather-extreme games. Fantasy projections handle weather worse than betting projections do, and we are open about that.
For everything else — usage-driven, schedule-driven, regression-driven cases — we will take the model over consensus every time, and the backtested hit rates back that up.
Bottom line
Consensus rankings are a useful baseline because they are everyone else's baseline. The edge comes from knowing where, specifically, your projection model disagrees and why. Trust full-tier and two-tier deltas, treat half-tier deltas as noise, and always read the contributing-feature breakdown before you act. Edges this small compound across an 18-week season into the difference between a playoff bye and missing the playoffs.
Open the Sharksnip fantasy rankings to see today's biggest model-vs-consensus deltas, sorted by position and tier gap.