Definition

A team treating a growth lever as causally connected to the commercial outcome it appears to drive, without running the test that would confirm the connection — and building further decisions on top of it.

Sincere steering, no connection

In The Simpsons, Maggie sits behind a toy wheel clipped to her car seat. It mirrors Marge's real one. Maggie turns it, the car turns when Marge turns hers. The correlation is perfect and continuous, but Maggie's wheel connects to nothing.

The provider-side version is a team operating its own lever — a proxy, an intervention, a "this is what works" belief — with the same earnest disconnection. The commercial outcome materializes, the lever-turning coincides with it, and the team credits the lever. No one runs the disconfirmation that would show whether the lever steers or whether something else is moving the car.

This separates the pattern from growth theatre and evidence theatre. Those perform a form — the form of growth work, the form of evidence-gathering. A team steering a toy wheel performs no form; the steering is sincere. The missing piece is verification that the wheel connects. The failure is operating an unverified causal claim as though it were confirmed.

The consequences compound

The single mistaken belief is cheap. What makes the pattern expensive is everything built on top of it:

• Compounding on unverified causation: every later decision built on "this lever works" inherits the unchecked assumption. If the wheel is disconnected, the whole downstream structure is steering on nothing.
• Investment escalation: resourcing, roadmap, and team structure organize around a lever whose connection was never confirmed. Under pressure, the team turns the wheel harder rather than checking the linkage.
• Attribution lock-in: when the car moves — driven by the user's own intent, not the lever — the team credits the wheel. The credit reinforces the behavior and makes the claim progressively harder to question.
• Opportunity cost: attention spent on the toy wheel is attention not spent finding the lever that does connect, or noticing the outcome was driven by something the team doesn't control.

Signal versus lever

A behavior that correlates with the outcome on its own isn't necessarily a lever that produces the outcome under intervention. A feature-usage event can track the outcome because high-intent users happen to perform it, not because performing it causes anything. Steering a toy wheel is declining to test that distinction — operating the lever without running the hold-test that would reveal whether it connects. The test is the remedy: an intervention amplifies the behavior, and the conversion rate among performers either holds steady (a lever) or drifts (hold failure — a signal). A value milestone is the lever that has passed this test repeatedly; a toy wheel is one that has never been put to it.

Example

A team notices that signups who connect a second data integration during onboarding convert to paid at three times the base rate. They read the correlation as a lever and build onboarding to push the second integration — a prompt, a checklist item, a nudge. The share of signups connecting a second integration climbs. Paid conversions climb too. The team credits the integration push and resources more of it.

The wheel may be connected to nothing. Connecting a second integration could be something high-intent buyers do on their way to converting — a marker of users who were going to pay regardless — rather than a step that moves anyone toward paying. The conversions might have climbed because the cohort's intent ran higher that quarter, or because a separate change moved them. The team never checked whether the conversion rate among second-integration users held once the intervention inflated their number. Without that check, "the integration push works" is a wheel the team turns earnestly, building roadmap on top of it, while the car moves for reasons it hasn't confirmed.

See also

• Proxy failure
• Multiplier analysis
• Model-level reasoning
• Correlation drift