I went hunting a cliff in my own rankings. The big, fast backs don't have one.

What the consensus says

The group chat has a comfortable story about big, athletic running backs: they age gracefully. The logic sounds airtight. A 225-pound back who ran a 4.4 has the hands to slide into a third-down role when the burst fades, so the cliff that wrecks ordinary backs at 28 gets a guardrail. FantasyPros' dynasty value pieces lean on pass-catching as the longevity tell, flagging the receiving backs as the ones who "stay on the field late." Footballguys frames the elite athletes as the holds worth keeping through the back nine. And everyone points at Derrick Henry, 31 and still leading the league in 20-plus-yard runs, as living proof the freaks get an extension. Buy the size-and-speed back, the story goes, because Father Time files him under "later."

The claim, in plain English

Here is the part that makes this personal. My own engine half-believes it. SignalTuned carries an archetype called dual_phase_back, heavy plus elite speed, and it hands that archetype a slower-decline bonus on the explicit theory that these guys "convert to pass-catcher as athleticism fades." So when my outcome-anchored error report flagged dual_phase_back as a place the engine over-rates players versus what actually happened, the obvious suspect was that bonus.

def predicate(row) -> bool:
    """In-cohort = dual_phase_back: combine weight >= 215 lbs AND speed_score >= 100."""
    return row["weight"] >= 215 and row["speed_score"] >= 100

(speed_score is Barnwell's old measure, weight times 200 over the 40 time to the fourth, so a big man who runs fast scores high.) The control is the whole game. These are the better athletes, so they post bigger seasons. I cannot ask "do they score more," I have to ask "do they fall off faster once the peak window closes." That means banding by age and measuring how much of last year's per-game line each back keeps.

How I beat on it

scripts/h57_dual_phase_back_post_peak_cliff.py builds Y to Y+1 pairs from player_stats_season.csv (REG, RB, 2015-2024 Y0, Y+1 reaching 2025, at least 6 games and 80 touches so the role is real), joins the combine for weight and 40 time, and joins birth dates for age. n = 438 pairs, 114 of them dual_phase_backs: Saquon, Jonathan Taylor, Henry, Breece Hall, Fournette, Pacheco. The metric is retention, next year's ppg divided by this year's, read inside post-peak age bands, because a decline-rate question is not a points-level question. Half-PPR. Full framework: the Session 97 two-gate spec.

What the data actually said

First swing looked like a smoking gun. In the age 29-to-31 band, the big, fast backs kept 81 percent of their per-game scoring while the field kept 98. A 17-point retention gap. Sell the freaks, the wrench is already out.

Because aggregates lie and small samples lie louder, I poked it. The whole gap was riding on one field back with a tiny prior-year line and a fluke bounce-back, the low-denominator ratio that detonates an average. Put a floor on it, only backs who actually scored the year before, and 17 points collapses to 7.

Then the cleanest control settled it. Fit the league's own next-year line, then ask how far below it each group lands once past the peak window. The big, fast backs come in at 1.49 points a game under the line. The field comes in at 1.33 under. The difference is 0.16 a game. That is not a cliff. That is the same staircase everyone walks down, and the freaks are standing on the normal step.

What the engine already figured out

So I went to yank my own dual_phase_back bonus, and the data told me to put the wrench down. The bonus assumes "these backs age slower," and they don't. But they don't age faster either, which is the only thing that would justify a penalty. Post-peak they decline on the ordinary running-back schedule. Henry is not the proof of a rule, he is the roughly 6 percent of peak seasons that land after 28, the outlier the story got built backward from.

The error report was not lying, though. The engine does over-rate this group. It just isn't an aging problem, so the aging knob is the wrong tool. The real leak sits upstream, where a freak combine line (the 220-pound burner who tested like a first-rounder and never earned the carries) collects credit the production never repays. That is a board-placement question, not a decline-curve one, and it goes on the bench for its own look.

What to do about it

Don't sell a big, fast back because you are scared of the cliff. He declines like every other back, no quicker. And don't pay the longevity premium either, because the "converts to pass-catching" insurance policy mostly doesn't pay out, it just doesn't cost extra. The athletic profile tells you how good the back is, not how long he lasts. No engine change ships from this one. I came to cut a number, and ten years of cohort talked me out of it. The boring answer won: the freaks age like the rest of us.

← Back to all posts