In 1999, Donella Meadows — co-author of The Limits to Growth (1972) — published a short essay, Leverage Points: Places to Intervene in a System. Drawing on three decades of system-dynamics modelling, she presented twelve places to intervene in a system, ranked from least to most powerful. The least-powerful were parameters (taxes, subsidies, standards) — what most policy debates fight over. The most-powerful were paradigms (the shared mental models a society operates from) and the power to transcend paradigms. The hierarchy was deliberately counter-intuitive, and Meadows's punchline was deflationary: the higher the leverage, the more the system will resist the change.
Meadows's hierarchy reads as three broad bands. At the bottom: fiddle-with-the-knobs interventions — parameters (tax rates, subsidies, standards, minimum wages), the size of buffers (strategic petroleum reserves, central-bank reserves, grain stockpiles), and the stock-and-flow infrastructure (highways, power grids, water systems). Important but rarely transformative; once the pipes are laid, they shape what is possible, and political debate has limited reach into what the system can become.
The middle band is where systems acquire their dynamics: the delays in feedback loops (long delays cause oscillation, overshoot, collapse), the strength of negative feedback relative to the disturbances it must correct, the gain on positive feedback (which amplifies and so destabilizes), the information flows that determine who knows what (adding monitoring often dramatically changes behaviour), and the rules of the system — property law, tax codes, antitrust, corporate governance — that set the incentives inside which the rest operates.
The top band is where most policy debate does not reach. The power to self-organize structure — the system's ability to grow new rules, new variables, new feedback loops — is paradoxically more powerful than any specific intervention because it lets the system find its own interventions. Above that sit the goals of the system (changing what an organization is trying to do usually changes everything else), the paradigms it operates from (growth is good, land can be owned, money is wealth — usually invisible to the people inside them, which is why they are so high-leverage), and finally the power to transcend paradigms. The pattern is consistent: lower-leverage interventions are easier to identify and make; higher-leverage interventions are harder to make and easier to miss, and the system resists high-leverage intervention precisely because high-leverage points are load-bearing for its identity.
Climate policy is the framework's clearest contemporary application. Carbon pricing (a parameter at level 1) gets enormous political attention; paradigm change — rethinking the relationship between economic activity and well-being — gets very little, and the hierarchy explains why the easier intervention is the politically dominant one. AI safety runs the same diagnosis: tweaking parameters (loss functions, reward models) is easier than tweaking the goals of the field or the paradigms under which systems are built. Personal development fits the same pattern, with most effort focused on parameters (calorie counting, time tracking) when the higher leverage may sit at goals or paradigms. Use the hierarchy as a diagnostic checklist: where in the hierarchy is the proposed intervention, and where is the problem? When those don't match, the intervention will fail in a recognizable way.