This is still forming, and it is the load-bearing one — so read it with the most care and the least credence.
Start from the question of what actually bounds the reachable. The work has already moved the wall off energy: Earth receives solar exergy at ~175 PW and civilisation dissipates ~18 TW, so the thermodynamic ceiling is distant (binding-constraint). Atoms are abundant too. What is scarce is knowing which arrangement of available atoms would expand the reachable set, and being able to assemble and coordinate it.
Distinction Physics gives this a precise spine. Every change of configuration is a trajectory through a space of distinctions, and its cost is the action S = ∫E dt — energy integrated over time, the same quantity classical mechanics calls action (DP, Module 4). A desired configuration is reachable when some feasible trajectory of bounded action reaches it. So the cost of reaching anything is action — energy × time — and "this problem is hard" turns out to mean only "this trajectory requires a great deal of action." Hardness is a magnitude, not a category. Nothing a configuration's difficulty alone can do puts it outside the possibility space.
What does put a configuration outside it is a physical invariant — a wall physics forbids crossing at any action (a conservation law, the entropy floor, an irreversibility on the relevant horizon). Those are the only true walls, and binding-constraint already names them. Everything else we call hard, scarce, unprofitable, or unreachable is a question of how much action a goal needs, whether anyone can find a low-action path to it, and whether anyone mounts the action at all.
Here is where intelligence enters, in DP's own terms. In physics, nature selects the least-action path for free — δS = 0 is a law of motion. But among configurations that do not yet exist, nothing selects it automatically; the low-action path has to be searched for. That search is what pattern-intelligence is: the faculty that approximates the least-action trajectory to a desired configuration — doing deliberately what nature does for nothing. It does not repeal the action cost; it finds the short way through, so the same goal costs far less action than blind search would.
So the framing, offered as direction rather than a settled law: the possibility space yields to action × pattern-intelligence. Action (energy × time) is the fuel; pattern-intelligence is what spends it on the right path instead of wandering — DP's "excess action" is exactly the wandering intelligence removes. Read this way, a step-change in pattern-intelligence — machine pattern-intelligence at scale — is potentially the largest lever we have ever had on the reachable set: tools that improve thinking that build the next tools, each round finding lower-action paths the last could not (configuration-generates-configuration, run forward and faster).
Two edges keep this honest, and they belong inside the claim, not after it.
First, it relaxes the search, assembly, and coordination constraints — not the physical ones. The sinks, the specific materials, the biosphere, entropy itself stay exactly where binding-constraint put them, and pattern-intelligence aimed at throughput presses harder against them, not softer. The lever is real; the floor is unchanged.
Second, the asymmetry cuts both ways. A bigger lever moves the world faster in whichever direction it points, and option-space destruction is fast and local while expansion is slow and systemic (asymmetry-of-option-space-change). Pattern-intelligence aimed at a configuration that is itself shrinking the reachable set accelerates foreclosure. By default — unaimed, or aimed at yield under a price that cannot see the sink — that is the cheaper direction. So this is better called the greatest available amplifier than the greatest lever: it makes the question of aim (coordination-as-move-evaluation) and of objective (viable-objective) more urgent, not less. In the terms of infinite-game, it is a far more powerful way to either keep the game open or end it.
That is also why money is not the thing. Money is neither action nor pattern-intelligence; it is a claim on them, and it converts to mounted action only where coordination and incentive carry it there. For a configuration that is unprofitable to assemble or whose pieces are held apart, the claim never converts — which is why unlimited willingness-to-pay can sit beside a physically-permitted cure whose low-action path no one ever traverses. The path is not forbidden and may not even be especially long; it is simply never taken.
The sharpest objection, shipped with this claim: that every general-purpose technology was, in its moment, going to end scarcity — steam, electricity, the Green Revolution, nuclear, the internet — and each relaxed a specific constraint only for the constraint to relocate, often to an irreversibility (aquifer drawdown, nuclear waste, attention capture, data-centre load). The base rate for "this one finally escapes scarcity" is poor. And the lever has a footprint: under the prevailing aim — yield under a price that cannot see the sink — a cheaper way to produce is first of all more production, so the rebound loads the binding constraints before it relaxes anything. The honest answer concedes both, which is why the claim is narrowed to relaxing the non-physical, arrangement-bound constraints that have actually been binding, while the physical floor stands and is pushed harder by default. That pattern-intelligence acts on the configuration-finding process itself may make it different in kind — but that earns a wider error bar, not a presumption of rescue. Hence forming, and hence amplifier rather than lever.