The physical-freak RB looks like a bargain right now. Two years of data agree. Seven years do not.

What the consensus says

Two weeks ago I wrote about Bijan Robinson entering his dynasty prime. RotoBaller reported it directly: 2,298 scrimmage yards across three seasons, a proven receiving role, elite athleticism. FantasyPros' dynasty startup values put him in the top tier with Jonathan Taylor and Bijan's Falcons teammate. Sharp Football Analysis frames the elite athlete at RB as a rare "safety and upside" combination -- the idea being that a player like Robinson, a 215-pound back who ran a 4.45, has more tools to age through a decline than the average bellcow. The big, fast back is the dynasty consensus darling right now. Not without reason.

But the same consensus runs on a story I tested two months ago and found thin. H-57 checked whether big, fast backs age slower than the field. They don't, at least not measurably. Post-peak retention gap: 0.16 points a game faster decline for the freaks. Which is to say, essentially nothing. So I went looking for a different channel.

The claim, in plain English

If the archetype premium is real, it should show up at the workload level: at equal season-level touches, a dual_phase_back RB produces more points the next year than a comparable back without the physical profile. The hypothesis is falsifiable.

def predicate(row: pd.Series) -> bool:
    """Cohort = dual_phase_back: weight >= 215 lbs AND speed_score >= 100,
    AND not classified as a receiving_back (touch_pct < 0.25)."""
    return bool(row["dpb"])

Speed score is Bill Barnwell's old metric: weight times 200, divided by the 40-yard dash time to the fourth power. A 220-pound back running a 4.42 scores 107. The receiving_back exclusion matters -- the engine classifies receiving backs first in its archetype hierarchy, so a big, fast back with a heavy target share exits the cohort. What's left is the bellcow bruiser with elite athleticism.

The control is weighted touches (carries plus half the targets). If the physical profile adds something the box score doesn't already capture, it should appear inside each workload band.

How I beat on it

scripts/h66_dual_phase_back_workload_premium.py runs Y-to-Y+1 pairs from the 2022-2024 weekly production file, aggregating to season level. 92 RB pairs total (minimum 8 games, 60 weighted touches). Fifteen qualify as dual_phase_backs: Derrick Henry, Saquon Barkley, Jonathan Taylor, Travis Etienne, Isiah Pacheco, AJ Dillon, Brian Robinson, and a few others. Framework: two-gate spec from Session 97. Half-PPR, Dynasdeez 10-team format.

What the data actually said

First swing: aggregate. Dual_phase_backs averaged 11.95 points per game in Year 0, and 12.44 in Year 1. The field averaged 10.29 in Year 0, and 9.93 in Year 1. That is a 2.5-point-per-game advantage for the freaks. The consensus is not wrong about this cohort right now. These backs are outperforming.

Aggregate first, bands second. Because aggregates lie.

The 200-299 band has the most signal: eight DPBs, 42 comparables, a 1.58-point gap that would clear the Gate 1 threshold except for one thing -- the DPB side has eight players, not ten. One back shy of the minimum-n rule.

More telling is the 300-plus band. True bellcow carries: three DPBs, two from the field. The freaks produce less. Derrick Henry's 2024 with the Eagles is the clearest case here -- 18.0 ppg, a field-defining year, but the data from his 2022-2023 seasons in Tennessee looked very different (9.6 and 11.2 ppg, below average for a 300-touch back). The point: the bellcow workload tier erases the DPB premium in this window, though n=5 total means you should hold that finding loosely.

Gate 1 verdict: fail. No band meets both the 1.5-ppg threshold AND the n-ten-per-side minimum. The positive aggregate is real, but it does not survive a proper within-band control.

What the engine already figured out

The engine carries rb_archetypes.dual_phase_back with one configurable lever: post_peak_modifier = 1.1, a slower-aging bonus. H-57 tested that lever on a ten-year sample and found the bonus is, charitably, not clearly wrong and, more precisely, is producing a 0.16-ppg overstatement of retention that falls well short of the Gate 1 threshold.

H-66 now runs the workload-efficiency test on 2022-2024 data and finds: no clean within-band premium. The two-year aggregate looks great because this particular window contains the surviving dual_phase_backs at peak -- Bijan entering his prime, Saquon's Philadelphia resurrection, Henry's Eagles run. That is a survivor sample. The players who declined out of the top-100 between 2018 and 2024 -- the Zeke Elliotts and Leonard Fournettes and David Johnsons who qualified as DPBs at peak and then fell hard -- do not show up in a 2022-2024 rolling window because they are no longer qualifying backs.

The seven-year outcome benchmark tells a different story. Across 2018-2024, when the engine put a dual_phase_back in its top 100, those backs landed an average of 47 rank positions lower than projected in realized multi-year value. That is a systematic miss. Two tests have now tried to find the mechanism. One found no faster decline. One found no workload-efficiency shortfall in the available window. The over-rating is real. The explanation is still missing.

What the engine is doing with the post_peak_modifier = 1.1: crediting a longevity story that H-57 says is not visible in retention data. What this test adds: even at equal workload in a favorable two-year window, the within-band premium is too noisy to stand on.

What to do about it

Two things. One is solid, one is a working theory.

Solid: do not pay a separate premium for the dual_phase_back label in dynasty trades. What you are actually buying when you trade for Bijan Robinson is his 2023-2024 production trajectory, his age-24 position on the curve, and his proven receiving role with Atlanta. Those inputs are already in the engine's projection. Adding "and he's a physical freak" on top of that is not supported by the within-band data -- the archetype label adds noise, not signal, once you've controlled for what he actually does.

Working theory: the seven-year over-rating may be injury-driven rather than production-driven. A 215-pound back absorbing 200-plus carries a year may produce more per game than a lighter back at the same workload but face a higher injury risk that cuts careers short. The realized multi-year value includes seasons lost to injury. If that is the channel, the engine's base projection is not miscalibrated on production -- it is undercounting the hazard rate. That question is logged for the next test.

Until that test runs: price the player. The physical freak sticker is not a separate line item. The box score already told you what he's worth.

← Back to all posts