WP-41 concluded the relocation is financially possible and politically implausible. This asks the question that leaves open — if we have the resources, why doesn't it happen? — and answers it with a model instead of a sentiment.
WP-41 priced partial relocation — 200 to 500 million people into high-altitude Americas refugia — at $20T to $100T over thirty to fifty years, needing $0.7–3 trillion a year sustained. The arithmetic closes: 200M × $100k is $20T, 500M × $200k is $100T, and the annual band contains both.
Put that band next to something the world already does. SIPRI puts world military expenditure in 2025 at $2,887 billion, being 2.5% of global GDP — which implies a world economy of about $115.5 trillion.
The top of WP-41's band is world military spending. The bottom is a quarter of it. Neither is a number civilisation cannot reach; both are numbers it reaches every year, for other things.
So the interesting question is not whether we can afford it. We can. The question is why that changes nothing.
Here is the mistake worth naming, because it is the one behind "if we have the resources we should just do it."
It assumes resources are the parameter — that a system's willingness to act is a function of how much is available, so that pushing availability up must eventually push action up. In a coordination problem that is false, and the falseness is structural rather than a matter of degree.
Each actor — a state, an institution, a fund — has a threshold: the fraction of others acting that would make acting worth it. Thresholds differ. Equilibria are the fixed points of
$$x = F(x)$$where $F$ is the distribution of thresholds, and an equilibrium is stable when $F'(x^*) < 1$. This is Granovetter's 1978 threshold model; nothing in it is new, and that is rather the point.
Now ask where resources enter. They do not enter as a level. They enter only by moving thresholds. A trillion dollars sitting in a place no actor's decision depends on moves no threshold at all, and the fixed-point structure does not know it exists.
Tangency of $F$ with the diagonal needs $F(x)=x$ and $F'(x)=1$ at once. With $z=(x-\mu)/\sigma$ that is $\phi(z)=\sigma$ and $\Phi(z)=\mu+\sigma z$, and $\sigma$ eliminates:
$$\Phi(z) - z\,\phi(z) = \mu$$Below the critical dispersion the system has a low fixed point. Across eleven parameter pairs, that fixed point sits between 0.0015% and 1.2% of actors acting — and $F'(x^*)<1$ at every one of them.
Read that carefully. The system is not moving slowly. It is at rest. Essentially nobody is acting, and that state is an attractor, and it will remain one however much money accumulates beside it.
That is the answer to the question. Abundance and paralysis coexist because they are properties of different variables.
Two ways to intervene. Guarantee: de-risk a fraction $g$ of actors outright so they act regardless, giving $F_g(x) = g + (1-g)F(x)$. Subsidy: make it cheaper for everyone, lowering every threshold by $d$.
Every intervention that tips the system is under 20%, against a programme that is 100%.
The instrument's job is not to fund the relocation. It is to move the system off a fixed point — after which the remaining actors move because it has become rational for them to, and they bring their own money.
∎Before computing any of this, I asserted — in conversation, confidently — that the guarantee channel would be about five times cheaper than the subsidy channel. The reasoning felt solid: de-risking a handful of first movers must surely beat subsidising everybody.
It is wrong. Across all eleven parameter pairs the ratio $d^*/g^*$ runs from 0.863 to 0.972, median 0.937. It is below one everywhere, which means the subsidy channel is cheaper — by three to fourteen percent. The estimate was wrong by a factor of about five and wrong in direction.
What survives is not nothing, but it is not what was claimed. The case for a guarantee instrument does not rest on cost-efficiency, and this paper withdraws that. It rests on something the model does not contain: a guarantee is contingent, so it is authorised against capital that is never spent. That is a fact about political economy, not about economy of resources — and it is a real argument, just not the one that was made.
WP-41's correction of 15 September withdrew the claim that receiving capacity does not exist — the world absorbed 1.52 billion international arrivals in 2025 and was paid to do it — and replaced it with four binding constraints: duration, livelihood, financing, political consent.
This paper addresses financing, and concludes it is not binding. It says nothing whatever about duration or livelihood.
On political consent it says only this: consent has the shape of a coordination problem with a stable stuck equilibrium. That is a description of the difficulty, not a route through it. No mechanism here delivers consent. A paper claiming otherwise would be repeating the exact error that the September correction already had to fix once.
This is the section the rest of the paper needs, and it is a warning about the rest of the paper.
A model that explains why nothing happens is one short step from a model that excuses it. "The low equilibrium is stable" can be read as a finding or as a shrug, and the difference is not in the mathematics. Every sentence above is compatible with using it as a reason to stop.
A guru levitates. Everyone gasps. He tells everyone to calm down — this is absolutely normal, he says. He levitates out of there, levitates away. Someone should tell him he could hurt himself playing like that. He levitates so high he disappears into the sky, and then he tells everyone who is out there looking for him: do not fear. It is absolutely safe.
Death is absolutely safe.
Every statement the guru makes is true. That is what makes it work. Nothing bad does happen to the dead, and his serenity is perfectly well founded — it is just that it scales with his altitude, and the people it is addressed to are on the ground with their necks craned.
Read §3 again with that in mind. The model puts more than 98% of actors at rest and calls the state stable. It is stable. $F'(x^*)<1$ is not in dispute. And calm down, this is absolutely normal is a fair summary of what the mathematics says, which is exactly the problem: the gasping is the only part of the system that has registered that something is wrong.
So: the object is prior to the account of it. The people in WP-41's table are there whether or not $F'(x^*)<1$. A white dwarf was doing what white dwarfs do for the whole history of the universe before 1930, and the calculation on the boat to Cambridge did not bring it into being.
Which is worth saying precisely, because the man who did that calculation ran into the opposite failure and it cost him thirty years. Eddington had a physical intuition — a star cannot just collapse, that is absurd — and used it to overrule a correct piece of arithmetic in public. The error here would be the mirror image: using a correct piece of arithmetic to dismiss a physical fact.
Both are the same mistake. Both let one side of the relation between the map and the ground overrule the other, and the direction only determines which way you end up wrong.
So the honest summary is narrow. The financing is not the obstacle, and the intervention that would move the system is smaller than the programme by an order of magnitude. That is what the model shows, and it is genuinely useful, and it is not a plan, and it does not touch the hard constraint, and none of it is a reason to be at rest about people who are not.
Seven gaps in the script. The one that most needs stating in public: $\mu$ and $\sigma$ are invented. No threshold distribution was measured. The numbers — 3.7%, 16.8% — are properties of chosen parameters, not estimates of anything in the world. What is robust across the sweep is the structure: a stable low equilibrium, a saddle-node, and a tipping intervention far smaller than the programme. Quoting 3.7% as though it were a measurement would be the worst possible use of this paper.
A second problem, which is worse. The figure is denominated in a currency whose meaning is conditional on the scenario not arriving. Cash buys nothing once the infrastructure that would sell you something is gone. A displacement large enough to need this programme is large enough to damage the order that gives a dollar figure meaning at the far end — so the price is well defined only in scenarios mild enough that it does not matter. Every affordability argument in §1, including the comparison to military spending, inherits that condition.
And the per-person relocation cost of $100–200k is inherited wholesale from WP-41. Every dollar figure here rests on it, and it was the least examined number in the chain.
It has since been corrected there, on 18 September, and the correction cuts under this paper too. $100–200k is not what moving costs — people move themselves for a few hundred to a few thousand, and 7,904 died or disappeared doing it in 2025. It is what a destination costs. Pricing relocation that way assumes the politics of arrival have already been won, which is precisely the variable §§3–5 of this paper say is stuck. The band $0.7–3T/yr should therefore be read as the cost of arrival at a standard, not the cost of movement, and the affordability argument in §1 is an argument about the expensive end only.
Producing script: book6/wp126-not-the-parameter-verify.py — 7 sections, 7 gaps. CITED: SIPRI, world military expenditure 2025; Granovetter, Threshold Models of Collective Behavior, AJS 83 (1978). Cost figures inherited from WP-41. Related: WP-42 (carrying capacity), WP-43, WP-44 (the manifold), and Book VII · Chandrasekhar, whose bifurcation is the same shape and whose Eddington episode is §7's warning.