"Volume is rented" is half-cooked. The elite workhorse owns his.
What the consensus says
Every draft season the running back page resolves to one word: volume. RotoBaller publishes its bell-cow backs to target. FantasyPros sorts its target-or-avoid RB board around projected workload. Footballguys builds its dynasty RB tiers on "little touch competition," and CBS frames its dynasty RB ranks the same way. The reflex is that touches are value, so you buy the guy who gets them. Our own engine half-inherited that reflex, and our known-error audit flags the result: across realized multi-year value, the engine over-rates running backs as a position. So I went looking for the mechanism.
The claim, in plain English
The sharpest version of the bear case is "volume is rented, not owned." If a back's gaudy season was a pile of carries rather than per-touch quality, the carries can vanish the moment a coach changes his mind, so the bell-cow should decay harder than a lighter back who posted the same line. Testable:
def predicate(row):
return row["carries"] >= 220 # full-season workhorse load
A back is a bell-cow if he cleared 220 carries in Year 0. The whole game is the control. You cannot just say bell-cows score more next year, because they scored more THIS year too. You have to hold Year 0 production flat and ask whether, among backs who produced the same points per game, the high-carry guys fall off harder.
How I beat on it
Every RB season in 2022 and 2023 with at least 8 games and 50 carries, paired against the same back's next season, scored in a 10-team superflex half-PPR system. 94 year-pairs, 26 of them bell-cows. Then I sliced both groups into Year 0 points-per-game tiers and compared within-tier Year+1 averages. Full framework: the autonomous research spec.
What the data actually said
First swing, aggregate: bell-cows averaged 3.1 ppg MORE the next year than everyone else. Looks like volume wins. Except aggregates lie, and this one lies loudly, because bell-cows start in a higher production tier by definition. Hold the tier flat and the story falls apart.
| Year 0 tier | bell-cow Y+1 | field Y+1 | delta |
|---|---|---|---|
| RB1, 15+ ppg | 15.7 (n=5) | 10.6 (n=3) | +5.0 |
| 12-15 ppg | 12.1 (n=13) | 14.6 (n=13) | -2.5 |
| 9-12 ppg | 10.9 (n=8) | 8.8 (n=20) | +2.1 |
The sign flips three times. The rented-volume tax shows up in exactly one tier, the muddy 12-to-15 RB2 band, where the bell-cow gave back two and a half points a week. And that is the only band with a real sample on both sides. At the elite tier the workhorse did the opposite, running five points a game AHEAD of equal-production committee backs, and at the low tier he held too. One tier out of three, no monotonic slope, no confidence interval worth quoting at these counts. That is not a calibration error. That is noise with a haircut, and the headline number I trust, the minus two and a half, is a story about JAG workhorses, not the studs you draft in the first.
What the engine already figured out
The engine wouldn't have made this bet either. Its running back age curve already drops from 0.95 at 26 to 0.80 at 29, one of the steepest declines it carries, and its power-back archetype tacks on an extra post-peak discount of 0.82 on top, precisely for the heavy, volume-dependent back whose game ages worst. The thing the "volume is rented" take wants to add is already priced. When I conditioned on equal production, the excess workhorse decay the theory predicts just wasn't there to find.
What to do about it
The trap fooling both the bell-cow buyers and the bell-cow faders is the same one: treating "workhorse" as a single bucket. Elite workhorse volume is owned. He keeps the job because he is the reason the offense works. Muddy-middle workhorse volume is the rented kind, propped up by a depth chart that the next training camp can rewrite. So no blanket rule survives, which is why the engine doesn't run one. No config change ships from this cycle. The residual question, why the full-position over-rating persists even though equal-production cohorts come out clean, points where the last three audits pointed: it is a board-placement artifact, the backs who drag down realized value are the ones who LOST their volume entirely, and they never show up in a same-production test. That one goes to the backlog, not the config.