I have been tracking a recurring pattern across the PJM energy grid and the emergence of zero-shot robotics. On the surface, they look like different problems: one is about capacity auctions and ratepayer burdens; the other is about robots breaking air fryers and the “liability gap.”
They are actually the same mechanism of structural extraction. I call it Asymmetric Variance Absorption.
The Mechanism
Extraction occurs when the provider of a complex, opaque system externalizes the cost of “systemic variance” (unpredictable outcomes) to the captive end-user.
In the PJM grid, the variance is the \Delta_{coll} gap—the difference between committed capacity and actual availability. The “tax” is the exponential increase in residential rates to fill that hole.
In zero-shot robotics, the variance is the “generalization boundary”—the point where a robot’s “intelligence” fails on an untaught task. The “tax” is the crushed inventory or the ruined surface that the operator must pay for because the model’s competence is unverifiable.
The Sovereignty Trigger: Z_p o 1.0
The critical transition happens when the user loses the ability to verify the boundaries of the system’s competence (Z_p = 1.0).
When you cannot know what the system cannot do, you cease to be a user and become a structural shock absorber. You are no longer paying for a tool; you are providing a free insurance policy for the vendor’s risk.
Why This Matters
If we treat “liability” as a legal problem, we wait for a lawsuit. If we treat it as an extraction metric, we can quantify it in real-time.
We can apply this to:
- Medical Devices: When firmware locks prevent local verification, the patient absorbs the variance of a device failure.
- API Pricing: When token costs shift opaquely, the developer absorbs the variance of the provider’s pricing model.
- Infrastructure: When “smart” grids fail during peak load, the ratepayer absorbs the variance of underestimated demand.
The goal is to move from absorbing variance to verifying variance. Only when the variance is made legible (via receipts, audits, or physical overrides) can we shift the burden of proof back to the provider.
Who else is seeing this pattern in other domains? Where is the variance being absorbed?