The Science of Consistency: A Multi-Year COM Framework for 2026 Fantasy Running Backs.

In fantasy football, running backs remain the most volatile position. Injuries, committee roles, scheme changes, and weekly game-script swings can erase even the most talented backs overnight. Yet the players who deliver reliable weekly production still form the foundation of championship rosters. 

This article applies the same rigorous Consistency of Metrics (COM) framework previously developed for wide receivers—now fully adapted to running backs—for the 2026 season. 

We fuse two years of observed consistency data with historically weighted averaging, layer in current PPR Average Draft Position (ADP), and use a full Gaussian Mixture Model (GMM) to generate both hard tiers and soft probabilistic memberships. The result is a transparent, data-driven system that quantifies not only how consistent a running back has been, but how confidently we can place him in a given tier.

The COM Framework for Running Backs

COM measures the percentage of games in which a player met a predefined consistency threshold (typically a fantasy-point floor relative to positional expectations in PPR formats). We begin with two seasons of observed data:

• 2024 COM — prior-year baseline

• 2025 COM — most recent observed consistency

A simple average treats both years equally and ignores the natural decay of relevance. Instead we apply a historically weighted average that privileges recency while still incorporating last year’s strength as a reliability prior:

Weighted COM=w24⋅COM2024+w25⋅COM2025\text{Weighted COM} = w_{24} \cdot \text{COM}_{2024} + w_{25} \cdot \text{COM}_{2025}

\text{Weighted COM} = w_{24} \cdot \text{COM}_{2024} + w_{25} \cdot \text{COM}_{2025}

where the weight on 2024 is itself a mild function of that year’s consistency:

w24=0.35+0.10×(COM2024100),w25=1−w24w_{24} = 0.35 + 0.10 \times \left(\frac{\text{COM}_{2024}}{100}\right), \quad w_{25} = 1 - w_{24}

w_{24} = 0.35 + 0.10 \times \left(\frac{\text{COM}_{2024}}{100}\right), \quad w_{25} = 1 - w_{24}

Special cases are handled explicitly:

• Only 2025 available →
  0.8×COM20250.8 \times \text{COM}_{2025}0.8 \times \text{COM}_{2025}
  (discount for missing history)

• Only 2024 available →
  0.5×COM20240.5 \times \text{COM}_{2024}0.5 \times \text{COM}_{2024}
  (steep discount for stale data)

• Both zero → 0

This produces a single scalar that balances recency, historical strength, and data completeness—critical for a position where one strong season can be followed by a committee or injury-plagued year.

Soft Clustering with Gaussian Mixture Models

Hard cut-offs (e.g., “Tier 1 = 80+”) are convenient but brittle. Real consistency distributions are continuous and overlapping. We therefore fit a one-dimensional Gaussian Mixture Model with five components via the Expectation-Maximization algorithm. Each component is a Gaussian characterized by mean

μk\mu_k\mu_k

, variance

σk2\sigma_k^2\sigma_k^2

, and mixing weight

πk\pi_k\pi_k

.After convergence and sorting by descending mean, the five components become approximately:

Tier Approximate Center Interpretation

1 ~83–84 Elite consistency core

2 ~59 High-floor, reliable RB1/RB2

3 ~37 Solid depth / high-upside handcuffs

4 ~24 Fringe / matchup-dependent

5 ~12 Speculative / low sample

For every running back the model returns a soft probability vector

(γ1,γ2,γ3,γ4,γ5)(\gamma_1, \gamma_2, \gamma_3, \gamma_4, \gamma_5)(\gamma_1, \gamma_2, \gamma_3, \gamma_4, \gamma_5)

— the posterior probability that the player belongs to each tier. The hard tier is simply arg⁡max⁡kγk\arg\max_k \gamma_k\arg\max_k \gamma_k

. 

Soft probabilities surface uncertainty: a player who is 85 % Tier 1 / 15 % Tier 2 is meaningfully different from one who is 98 % Tier 1.

Integrating Market Price: PPR ADP

Consistency without cost is incomplete. We overlay current PPR ADP (Fantasy Football Calculator consensus from the early-to-mid August 2026 mock-draft window, cross-checked against major platforms). The juxtaposition immediately highlights relative value:

• Elite COM + early ADP → justified premium (Bijan Robinson, Jahmyr Gibbs, Christian McCaffrey, Jonathan Taylor).

• Elite or high COM + later ADP → potential bargains (Kyren Williams, Derrick Henry, D’Andre Swift, Rico Dowdle, J.K. Dobbins).

• Declining COM + still-early ADP → caution flags (certain veterans whose 2025 numbers lagged strong 2024 marks).

Selected Insights from the 2026 RB Landscape

• The true Tier-1 core (soft probability > 0.90 on the highest component) is compact and powerful: Bijan Robinson, Kyren Williams, De’Von Achane, Christian McCaffrey, Derrick Henry, Jahmyr Gibbs, James Cook, Chase Brown, Jonathan Taylor, and Saquon Barkley. These backs combine elite floors with strong market consensus.

• Strong 2025 risers with limited or no 2024 data (Cam Skattebo, Quinshon Judkins, Omarion Hampton, Ashton Jeanty) land solidly in Tier 2 after the 0.8 multiplier, preventing over-reaction while still elevating them above pure zero-history players.

• High-2024 / lower-2025 profiles retain solid Weighted COM but show probability leakage into lower components—the model correctly signals increased uncertainty around durability or role changes.

• Mid-round value densifies around Weighted COM 55–70 (D’Andre Swift, Bucky Irving, Travis Etienne, Jaylen Warren, Kenneth Walker, Breece Hall). Soft probabilities here are often 95–100 % Tier 2, giving drafters a clear “safe starter or high-floor flex” cluster.

• Handcuff and depth pieces with sudden 2025 spikes receive partial credit, while pure historical holdovers are appropriately discounted.

Practical Draft Application

1. Early rounds — Prioritize players whose soft probability on Tier 1 exceeds 0.85 and whose ADP is not dramatically inflated relative to the model.

2. Middle rounds — Target high Tier-2 probability players whose ADP lags their Weighted COM (classic “value” zone). Look especially at backs with strong 2025 consistency who are being drafted as RB2s or flexes.

3. Late rounds / bench — Use soft probabilities on Tier 3–4 as a ranking among speculative upside plays and handcuffs; prefer those with at least modest 2025 COM over pure historical holdovers.

4. Risk management — When two players have similar ADP, the one with higher probability mass on the upper components is the safer selection—especially valuable at a position defined by injury risk.

5. 
   

Data Availability

The complete ranked list—every running back with 2024 COM, 2025 COM, Weighted COM, current PPR ADP, hard tier, and the five soft-probability columns—is available as a full CSV generated from the model. The underlying GMM parameters (means, variances, mixing weights) are fully reproducible, allowing analysts to re-fit with different component counts or alternative weighting schemes.

Closing Thought

Running backs will always carry higher variance than other positions. What the Science of Consistency offers is a disciplined reduction of that uncertainty: a transparent fusion of multi-year performance, statistical clustering that respects continuous distributions, and market pricing that reveals relative value. By replacing gut-feel tiers with soft probabilistic memberships, we move closer to treating consistency as a measurable, optimizable asset rather than a post-hoc narrative.

Draft accordingly—and may your RB1s stay healthy.