Data / entry 12 / August 8, 2026 / 8 min read

Sorry Vegas, My NFL Rookie Model Disagrees With You

49 players, seven weighted components, scored against a live betting market. Full component heat map, the two-factor scatter the model actually leans on, model-vs-market divergence, and weight-sensitivity ranges below.

How the model is built

This is a structured prior, not a fitted model: none of the seven weights were estimated from historical outcomes, because camp-signal and fan-sentiment have no clean training set. The weights are the hypothesis; the sensitivity data below is the test of whether that hypothesis actually drives the ranking.

Full component heat map

All 49 players, all 7 components, on a 0-100 scale, sorted by composite rank. DC = Draft Capital, OPP = Opportunity, FIT = Scheme Fit, CTX = Team Context, CAMP = Camp Signal, MKT = Market Sentiment, FAN = Fan Sentiment.

Reading the rows: David Bailey (#1) is the only player who scores hot across essentially every column at once, talent, opportunity, and camp buzz all pointing the same direction, which the memo itself flags as rare. Compare that to Fernando Mendoza (#7), the No. 1 overall pick, whose row is a genuinely mixed heat signature: elite on DC and MKT (the market-facing, pedigree-facing columns) but lukewarm on OPP and CAMP (the on-field, right-now columns) because Kirk Cousins is still the starter in Las Vegas.

The two-factor view: Opportunity vs. Camp Signal

The two largest weights in the model (30% and 18%) plotted against each other. Bubble size = composite score, shading by tier.

The upper-right quadrant, high opportunity and high camp signal together, is where the model's two biggest levers agree, and it's a short list: Bailey, Sonny Styles, Rueben Bain Jr., Caleb Downs. The lower-left-to-upper-middle band is the more interesting read: players like Ja'Kobi Lane and Kahlil Benson carry real camp signal or opportunity with zero market price behind them, which is exactly the gap between what sportsbooks can see and what a beat-reporter-level model can see.

Model vs. market: where the priced players diverge

Divergence = model z-score minus market z-score, computed only across players with a live award price. Positive = model higher on the player than the market.

This divergence is a shrunk estimate by construction: market sentiment is both a model input (10% weight) and the benchmark it's compared against, so real disagreement is understated here, not overstated. The Market-blind sensitivity regime removes that circularity.

Weight sensitivity: which rankings are real, and which are artifacts

Each player re-scored under 5 weight regimes (Base, Opportunity-led, Pedigree-led, Camp-led, Market-blind). Rank range = worst rank minus best rank across regimes.

David Bailey ranks #1 under all five weight regimes, a rank range of zero, the definition of a robust call. Fernando Mendoza, by contrast, ranges from #2 (Pedigree-led, where draft slot and market sentiment dominate) to #16 (Camp-led, where his lukewarm on-field signal drags him down). 19 of the 49 players in the full model show a rank range above 10 places, meaning for more than a third of this board, the specific rank is more a function of the analyst's chosen weights than of the underlying evidence.

Putting it together

Four independent lenses on the same 49 players converge on one honest conclusion: the model's high-confidence calls are the ones where opportunity, camp signal, and market price all point the same direction at once (Bailey, Styles, Downs), and its lowest-confidence calls are exactly the ones the sensitivity data flags as weight-driven, not the ones the composite rank alone would suggest. The heat map shows where the evidence clusters; the scatter shows where the model's two biggest levers agree or disagree; the divergence chart shows where the model breaks from the crowd; and the sensitivity chart is the honesty check on all of it. That's the actual shape of a defensible pre-season model: not a single number per player, but a stack of views that either reinforce or undercut each other.