Clutch Gene or Statistical Ghost?

Drag the lineup's star off (or onto) the floor and watch every teammate's stat line move along a line it was already on.

Grounded in "Every Ablation Is a Dose: Counterweights and the Semblance of Self-Repair" (arXiv:2610.02173). The paper ablates components of a language model with signed strength λ and finds each remaining component's response is a fixed affine law, E_r(λ) = own_r + γ_r·λ, with a slope γ_r baked into the weights before any ablation happens. The post-ablation "bump" other components show isn't new compensating behavior kicking in — it's the same straight line, evaluated at a different λ. This toy swaps "component" for "teammate" and "ablation" for "the star sitting / getting extra usage."

λ — Star Ablation Strength

0 = star plays normally · 1 = star fully removed · −1 = star given extra usage (the law runs both directions)

λ = 0.00

−1.0 (star overloaded) 0 (baseline) +1.0 (star fully out)

Points Across the Full λ Range

Each teammate's full affine line E_r(λ) = own_r + γ_r·λ, drawn before you even touch the slider.

Player	own_r	γ_r (slope)	E_r(λ) now

Reading the law: every bar on the left and every dot on the right is computed live from the same fixed numbers — own_r and γ_r per player — plugged into one formula. Nothing "decides" to compensate at λ=1; the value at λ=1 was determined by γ_r the moment the roster was set, same as the value at λ=0. Prop market angle: a book pricing "Role Player X's points O/U if Star sits" is implicitly estimating exactly this γ_r — a fixed sensitivity to the star's usage, not a live read on who's "stepping up" emotionally. If the paper's finding holds for neural network components, it holds even more cleanly for role players whose box-score correlation with a teammate's absence has been visible in the data for years: the "bump" is priced before the ball is even tipped.