ROLLBACK LAB

The margin for
human control.

Test whether a rollback package holds below a fixed frontier reference, or leaves a route back to recursive capability growth.

THE QUESTION

Can retained technology stay below the point where it regenerates what we intended to prevent?

Inspect the model ↗

SIMULATED TRAJECTORY

Recoverable capability

Index · frontier reference = 100Horizon

Central assumptions 27-case stress envelope Reference threshold

RECONSTRUCTION ROUTES

The weakest barrier matters.

Each route is tested against the strongest modeled actor. Below 1× means the required reserve is missing; below 1/M means resources meet the assumed reconstruction cost.

THE SAME SCENARIO, IN HUMAN TERMS

Human outcomes

See this exact scenario in the story ↗

Conditional risks and stopping assumptions

Rates apply per consequential transition. Reconstruction cycles measure a separate process.

Risk behavior Display horizon Consequential transitions Transitions for each hazard to halve Residual risk, percent of starting hazard Transitions before effective closure Assume every material reconstruction route stays closed. Assume indispensable human control remains in machine production and improvement. Assume essential services, liberty, and reciprocal human roles remain viable. Assume the warning is detected while humans can still contain the systems.

02 / COMPARE & CHALLENGE

Find where the conclusion changes.

Compare packages, change two controls at once, and see which assumptions matter most.

TWO-CONTROL SWEEP

Stability landscape

121 packages

Horizontal Vertical

Candidate stable Sensitive Regenerates Above enforcement ceiling

Every cell includes all 27 stress cases. Select a cell to apply its controls. Other inputs stay fixed.

ONE CONTROL AT A TIME

What moves the margin?

±10% of range

Changes in the worst resource margin, holding every other input fixed. A wider span means greater local sensitivity.

Shown as log₂ change from the current margin. Controls are clipped at their endpoints. This ranking does not capture every interaction.

03 / EVIDENCE & ASSUMPTIONS

An inspectable argument.

Mechanisms grounded in sources. Numerical judgments left open to challenge.

Sources checked
23 September 2026

MODEL SPECIFICATION · V1.0.0Equations, assumptions & limits+

Built from W360 Global Intelligence’s rollback framework

“A technological plateau for human sovereignty,” updated 13 September 2026. The model carries forward route-specific reconstruction margins, retained substrate, human technical authorship and an enforcement ceiling. It does not model essential-service continuity, political rights or transition harms as interchangeable scores.