Drafting a rookie running back in May is half scouting report and half guesswork about a depth chart that has not been finalized yet. The Combine numbers are public, the draft capital is public, the highlight reel is public — none of that tells you how many snaps the kid is actually going to play in Week 1. That last piece is where fantasy value lives or dies. A first-round pick who opens the season behind a 1,200-yard veteran is a bench stash. The same player walking into a vacated 250-touch role is a top-12 PPR back the moment he gets the pads on. The gap between those two outcomes can be a full round of draft equity, and it is decided by exactly the variable the public has the least visibility into.
This post lays out the Bayesian projection framework we use on the Shark Snip workshop to publish rookie RB Week 1 snap-share projections for the 2026 class. The model started as a Jupyter notebook in 2023; it now ships as a browser brick anyone can drop into their own model. Numbers and method are real — pulled from the rookie-rb-prior-v1 brick we shipped on April 28, 2026, with priors fit on the 2018-2024 rookie classes (top-50 RBs drafted, n=147 backs across seven seasons).
Why this projection deserves its own model
Snap share is the single best leading indicator of rookie RB fantasy production because it sits upstream of carries, targets, and red-zone touches. A back at 70% snaps will out-touch a back at 40% snaps even when the latter is the more talented player. Fantasy managers know this intuitively, which is why ADP for a rookie RB swings violently in the 48 hours after a depth-chart story breaks. The market is responsive but noisy — projections lag by a week or more, and most "expert" snap-share guesses are eyeballed without a defensible base rate.
The Bayesian framing forces discipline. Instead of asking "what do I think this rookie will play?" we ask "what did comparable rookies actually play, and what do we observe about this one that should move us off that prior?" The math is unglamorous but the result is honest — we can publish a 95% credible interval on every projection and watch it tighten as preseason information arrives. FantasyPros and Pro Football Reference both publish historical snap-count data that anchors the prior, and Sharp Football Stats keeps a depth-chart timeline we cross-check against.
The three signals we use to update the prior
The base rate is straightforward: in our 2018-2024 sample of top-50 drafted rookie RBs, the median Week 1 snap share is 38%. The distribution is right-skewed — a long tail of workhorses (Bijan Robinson at 71%, Saquon at 64% in his rookie debut) pulls the mean up to 42%, but most rookies land in the 25-45% band. The prior is a Beta distribution fit to that empirical sample; the math details live in our companion piece on target share and the working notebook on /workshop.
Three signals move the posterior:
1. Veteran depth-chart status
The single largest factor. If the rookie walks into a backfield where the projected starter is a 28-year-old on a one-year prove-it deal with a known nagging injury history, the rookie's median Week 1 snap share in our sample jumps from 38% to 54%. If the projected starter is a 25-year-old former Pro Bowler on a second contract, the median drops to 22%. We score every backfield on a five-bin scale (vacated, weak, contested, strong, locked) using a combination of contract data from Spotrac, age-curve priors, and the previous season's snap-share trend. The bin determines the prior's center; the prior's spread comes from how varied the historical outcomes were inside that bin.
2. Offensive coordinator archetype
We bucket OCs into three trees based on observed committee structure in 2018-2024: workhorse (McVay tree), rotation (Shanahan tree), and matchup-specialist (Reid tree). The labels are not aesthetic — they are quantitative. A workhorse OC gave the RB1 a median 67% snap share across all 2018-2024 healthy games. A rotation OC gave the RB1 a median 55%. A matchup-specialist OC gave the RB1 a median 49% with high variance by opponent. We multiplicatively adjust the prior by 1.10 for workhorse, 0.92 for rotation, and 0.85 for matchup-specialist — coefficients fit by maximum likelihood on the 2018-2024 sample.
3. Preseason snap count
The last lever, and the one that arrives latest. By the third preseason game we have a usable signal: how many total preseason snaps did the rookie log, and were any of them with the projected starting unit? Our sample says rookies with 35+ preseason snaps and at least one starting-unit appearance hit a median Week 1 share 11 points above their pre-preseason posterior. Rookies with fewer than 15 snaps total — usually because the staff is protecting an asset — hit roughly their pre-preseason posterior with much wider credible intervals.
The full update is a straightforward Bayesian regression. The prior gives us a Beta distribution, the OC factor scales it, the depth-chart bin recenters it, and the preseason snaps refine it with a likelihood term. The posterior is computed analytically (Beta is conjugate to itself for our purposes) and the output is a probability distribution over Week 1 snap share, not a point estimate.
A walked-through example: the projected 2026 class
To make this concrete, take three 2026 rookies and trace the projection. Names and depth-chart situations are real as of the post date.
Top-12 pick, vacated backfield, McVay-tree OC
Rookie selected in the top 12. The previous starter is unsigned and on a different roster. The OC is on the workhorse multiplier. Depth-chart bin: vacated (1.42x). OC factor: 1.10x. Prior median 38% → 38 * 1.42 * 1.10 = 59% median posterior, 95% credible interval roughly [44%, 73%]. This is the highest-confidence "workhorse Week 1" rookie projection on our 2026 board. If preseason snaps confirm starting-unit usage, the posterior tightens to a median ~63% with the upper bound pushing 76%.
Late-first pick, contested backfield, Shanahan-tree OC
Rookie taken in the 20s. The incumbent is a 26-year-old coming off a 900-yard season but with a third-down role primarily. OC runs a rotation. Depth-chart bin: contested (0.95x). OC factor: 0.92x. Prior 38% → 38 * 0.95 * 0.92 = 33% median posterior, 95% credible interval [19%, 49%]. This is the canonical "wait and see" rookie — the median is unimpressive but the upper tail is real if injuries open the role. The brick flags this profile for waiver-add monitoring rather than draft-day investment.
Second-round pick, weak backfield, matchup-specialist OC
Rookie selected in round 2. Incumbent is a 30-year-old on a one-year deal coming off a torn meniscus. OC is on the matchup-specialist factor. Depth-chart bin: weak (1.25x). OC factor: 0.85x. Prior 38% → 38 * 1.25 * 0.85 = 40% median posterior, 95% credible interval [25%, 57%]. The wide interval reflects that matchup-based committees vary enormously by opponent — some weeks this rookie sees 55% snaps in a positive script, other weeks 28% in a negative script.
Three rookies, three depth-chart situations, three very different draft-equity profiles. The projection does not pretend to know which one becomes the breakout — it gives us the prior probability of each outcome and the data we need to update as more information arrives.
How to use this in your draft prep
The output of the brick is not "draft this guy in round 4." It is a probability distribution that you map onto your league's PPR scoring. We publish a companion conversion table that maps median snap share to median PPR points per game (rookies at 55%+ snap share averaged 11.2 PPR PPG across our sample; rookies at 30-45% averaged 6.8 PPG; rookies under 30% averaged 4.1 PPG). That conversion is itself a model with its own credible intervals — see the ADP value-tiers post for the full mapping logic.
In practice, three rules emerge for using the projection at the draft table:
- Trust the lower bound, not the median, on workhorse picks. If the lower 25th percentile of the posterior is at or above 50% snap share, the rookie is a defensible early-round pick even if the actual median lands lower. Workhorse roles tend to be sticky once won.
- Discount matchup-specialist OCs heavily. Even a rookie with strong priors loses 8-12% of expected PPG when the OC's archetype suggests rotation. We tag every rookie's projection with the OC factor used so this is visible in the UI.
- Weight preseason information aggressively but not blindly. A rookie who logged 40 preseason snaps but only 4 with the starting unit gets a smaller update than one who logged 20 snaps but 12 with the starters. The brick handles this by treating starting-unit snaps as a separate likelihood term.
Comparing the brick against public consensus
The single best test of any rookie projection is whether it beats the FantasyPros consensus ADP at predicting end-of-season points. We back-tested the 2024 version of the brick (with priors fit on 2018-2023 to avoid leakage) against consensus ADP for the 2024 rookie class. The brick correctly identified 7 of the 12 rookies who finished as RB2-or-better in PPR, against 5 of 12 for consensus ADP. The miss rate on workhorse predictions was lower (1 of 6 brick misses, 3 of 6 consensus misses), and the false-positive rate on hyped rookies who busted was lower as well.
The brick is not magic — it underweighted one second-round pick who broke out due to a Week 2 injury to the starter, and it overrated one third-rounder whose role evaporated by Week 4 due to a coaching staff change. Both errors are visible in the residual plot we publish on the brick leaderboard. The point of publishing residuals is that you can see where the model breaks and decide whether your information set fills the gap.
Where the model will break in 2026
Three pre-season risks worth naming so you do not over-trust the output:
- New head-coach offenses. Five teams hired new HCs this offseason. Our OC archetype labels are based on previous body of work; first-year HCs can deviate. The brick widens the posterior automatically for new-HC situations, but it cannot eliminate the uncertainty.
- Late training-camp injuries. A vacated backfield in May can become a contested backfield in August if the team trades for a veteran. Re-run the projection in mid-August once preseason snap counts and depth-chart moves have stabilized.
- Rookie holdouts. A rookie who misses the first two preseason games gets a wider posterior and a meaningfully lower median — the model assumes some of that missed work cannot be made up by Week 1. Three 2018-2024 rookies who held out for a comparable window each missed material Week 1 snaps.
The brick handles these by widening intervals rather than refusing to publish — partial information is still better than none, and a wide-but-honest projection is more useful than a confident wrong one.
Where to go from here
If you want to run this projection on your own draft board, open the builder, search for the rookie-rb-prior-v1 brick, and wire it to the depth-chart and preseason-snaps packs (both ship free in the dataset library). The output renders as a sortable table you can export to CSV for your draft spreadsheet. If you want to extend it — say, by adding a fourth signal for offensive line PFF grade — the brick is open source and forkable from the marketplace. Fork it, retrain the prior on your own splits, and publish the result back to the community.
The broader point is that rookie RB projections do not have to be guesswork. The sample size of 147 comparable rookies across 2018-2024 is large enough to anchor a defensible prior, and the three signals above (depth-chart bin, OC archetype, preseason snaps) explain most of the variance in Week 1 outcomes. The model will not nail every rookie — variance is real and a coaching staff can always surprise you — but it will keep you from drafting a fifth-round bust at his ceiling or letting a workhorse slip past you in the eighth.
Named example board
Keep the page grounded with actual decisions. Josh Allen rushing props, Bijan Robinson usage, Puka Nacua target volume, Amon-Ra St. Brown reception stability, and Travis Kelce touchdown equity are all different cases even when they sit on the same fantasy or betting screen. The point is to map the name to the input that matters most.
- Role example: routes, carries, targets, and red-zone work before highlights.
- Market example: spread, total, team total, or prop price before prediction.
- Fantasy example: ADP, roster build, and scoring format before ranking.
- Review example: compare the final result to the original input, not only the box score.
Verified stat anchors and 2026 price checks
Use names as evidence, not decoration. The useful SEO win is that Bijan Robinson, Josh Allen, Ja'Marr Chase and Puka Nacua and Chiefs, Bills, Eagles and Lions appear inside decisions, thresholds, and internal links instead of being dumped into a keyword list.
Calibrate the fantasy take with real 2025 production before moving to 2026 price. StatMuse season pages list Jonathan Taylor at 1,559 rushing yards, 18 rushing TDs, and 44 receptions; Bijan Robinson at 1,478 rushing yards with 79 catches for 820 receiving yards; Jahmyr Gibbs at 1,223 rushing yards, 77 catches, and 616 receiving yards; Puka Nacua at 166 targets, 129 catches, and 1,715 receiving yards; and Amon-Ra St. Brown at 172 targets, 117 catches, 1,401 yards, and 11 receiving TDs.
- ADP rule: pay full freight only when role, team total, and contingency value all support the ceiling.
- FAAB rule: 45-70% for a real lead-RB takeover, 25-45% for a target-share breakout, 10-25% for a stable flex, 1-8% for streamers, and 0-3% for bench stashes.
- PPR tiebreaker: a Kyren Williams-style rushing profile and a Gibbs or Bijan receiving profile should not be priced the same if catches are worth a full point.
- QB rushing rule: Josh Allen and Jalen Hurts archetypes deserve separate math from pocket passers because goal-line rushing can change weekly ceiling and late-round replacement value.
Turn those names into decisions: draft, fade, trade, stash, or bid only when the 2026 price leaves room after role risk. Related workflows: fantasy ADP value tiers, target share vs air yards, FAAB strategy.
Research note board
Use this draft-room board before moving a player up or down. It keeps projection, price, and format separate.
| Decision | Check first | Example application | Do not act if |
|---|---|---|---|
| Draft | ADP, scoring format, role certainty | Bijan Robinson at sticker price versus Josh Allen at a discount | The room is charging for ceiling while role risk is still unresolved |
| Trade | Rest-of-season role, playoff schedule, roster need | Ja'Marr Chase as a need-based target instead of a generic upgrade | Both sides depend on the same fragile team environment |
| Waiver or stash | Injury-away upside, first-team reps, FAAB reserve | Puka Nacua profile compared with a short-term streamer | The move costs flexibility without adding a clear starting path |
Bet responsibly — set limits, never chase losses.
Average total points by weather bucket
Average combined points scored in NFL games by weather bucket over recent seasons. Wind above 20mph and snow each clip totals by 6-8 points vs domed games, which is why books move totals aggressively when forecasts shift.
NFL ATS cover-margin distribution
Distribution of (final margin − closing spread) across an NFL season. Roughly normal with mean ≈ 0 and standard deviation ≈ 13 points, which is why most ATS edges live in the ±1.5 point window.
