WP39 proves two theorems about a three-layer box model. T1: at fixed column burden $Q$, surface concentration is $C = Q/h$, exactly. T2: a 0/1 lid commutes with pointwise loss, does not commute with vertical transport, and therefore does not commute with a fold that carries transport. Order matters, and T2 says where the mattering lives.
WP40 built a curriculum on it. WP41 instantiated $K$ as a deployed ice-nucleation aerosol at 2–4 km and costed a planetary programme around firing it before the fold. On the evidence then available that was the obvious candidate and a reasonable one: it acts at roughly the right altitude, it has a threshold character, and unlike every other object in the frame it is something a person can actually release. WP66 §2 then argued that the instantiation does not satisfy T2's hypotheses. WP67 found that the deployment had already happened, unconsented and unmeasured, as atmospheric microplastic loading. WP68 turned WP66's four tests into a preservation filter.
Propose, develop, test, fail, turn. That is what the arc did, and it is what anyone does. Nothing in this paper is available to someone who has not done it, which is the point of §5.
WP41's own August addendum comes closest to the turn and then stops:
That withdraws one candidate. It does not ask what the class of candidates is — which is the only question that would have stopped the search instead of redirecting it. The arrow back was never drawn. This paper draws it.
WP39's T2(i) is proved in SmokeBox.lean for one specific loss: squaring, standing in for coagulation. The proof does not use the squaring. Inspect it and what it uses is two properties — the loss acts inside each layer with no term coupling one layer to another, and it sends an empty layer to an empty layer. Both are generic. So T2(i) is not a fact about coagulation; it is a fact about a shape.
WP39's T2(i) falls out as the case $f = (\cdot)^2$, and is re-derived in the file as a corollary to make the dependence visible. The contrapositive is the form a referee can run without knowing any atmospheric science:
The reason to want a $K$ was order. T1 pins the surface number to $Q/h$; T2(iii) says the lid and the fold do not commute, so when the lid fires relative to the fold changes that number. That is the only lever the model has. An operator that commutes with the gate has no “before”: applying it ahead of the gate and applying it behind the gate return the identical column, in every state.
This is not a claim that such an operator does nothing. An aerosol that removes mass reduces $C$; that is real and T1 prices it. The claim is narrower and harder to escape: such an operator cannot carry order-dependence, and order-dependence was the entire content of the framework being invoked. You may still argue for a pointwise intervention. You may not argue for it from WP39.
| candidate | what it acts on | commutes with the gate? | verdict |
|---|---|---|---|
| ice-nucleation aerosol, released at 2–4 km inside the mixed layer (WP41 as published) | phase of particles at a level — layer-local, and $f(0)=0$: no particles, no nucleation | yes, by K1 | not a gate |
| distributed microplastic loading (WP67) | the same slot, at unmeasured magnitude, with no off switch | yes, by K1 | not a gate, and an uncontrolled perturbation to the loss term |
| precipitation seeding | layer-local removal | yes, by K1 | not a gate |
| subsidence, frontal passage, boundary-layer cap, nocturnal inversion | $h$ — the denominator of T1 | no | gates, and not deployable |
| stratospheric aerosol acting on the radiation budget above the layer [MODEL] | $h$, indirectly, by changing surface heating — not layer-local | no | gate-shaped, and the one with the governance problem |
K1 takes four lines of Lean and its proof is two sentences. It does not follow that it could have been written in March, and this paper should not be read as saying so. It could not have been.
What WP39 established was T1, T2, and a real episode. What the $K$ slot then needed was an occupant, and nothing in T2 says which objects are eligible — T2 pairs the gate with the fold and says where non-commutativity lives. Turning that into a test on candidate interventions requires a specific candidate, at a specific altitude, developed far enough that its shape is visible. That is what WP40 and WP41 supplied. WP66 §2's observation — the aerosol is released inside the mixed layer, so it acts on phase and not on $h$ — is not available before somebody writes down a deployment at 2–4 km. And WP67's half cannot be reasoned to at all: that the thing had already been running for fifty years and had produced no effect anyone can measure is a fact about the world, found by looking.
So K1 is what WP66's observation looks like once it is stated for the class instead of for one candidate. Generalising it is a small step. It is small only after someone has taken the large one, and its whole value is prospective — the next candidate can be screened in a line, by someone who never has to run this arc.
Book 3, Chapter 3 carries a caution with the same algebra in it: “$K$ as a 0/1 gate and $F$ as a pointwise fold commute exactly. Non-commutativity in this framework comes from inter-site coupling inside $F$, not from the gate.”
That is not this screen, and it is worth being exact about why. The caution pairs the gate with the fold, and its target is a derivation — it exists to forbid a spurious boundary term $\propto \delta(\eta - \eta^*)$. K1 pairs the gate with a proposed intervention and quantifies over all of them. Getting from one to the other requires noticing that the candidate occupying the $K$ slot is itself layer-local, which is WP66's contribution and nobody else's. The algebra rhymes; the claim is different. Reading the earlier sentence as the later theorem is only possible once you have the later theorem.
Kepler published the nested Platonic solids in 1596 and was still defending them in the second edition of 1621, a quarter century later. The Astronomia Nova came out of the same work: he got the ellipse because he did the polyhedral programme thoroughly enough to reach an eight-arcminute residual in Mars against Tycho’s data, and then declined to discard the eight arcminutes. The lesson usually drawn is that he was wrong for twenty-five years. The more useful one is that the wrong model, worked honestly, is what produced the residual — and that what decided it was a measurement, not an argument. This arc turned in months rather than decades, and it turned for the same reason: WP67 went and looked.
One error on WP41 is in a different class, and the distinction is worth keeping. §3 assesses three receiving regions at 2–5M, 3–8M and 5–10M, then states a total of ~200–500M — roughly twenty times its own components, carried back from the scenario in §2 rather than computed in §3. It is corrected on the page.
The operator question needed WP66 and WP67 to close. This one needed addition. Both figures are on the same page, in the same section, and neither depends on anything learned later. The two errors are not the same kind of thing and should not be filed together: one is the cost of finding out, and the other is a section nobody re-read because the interesting claim was elsewhere.
An external reading of the arc, supplied to the author this week, argues that the interesting content of $K$ is political: that $K$ marks the point at which investigation becomes commitment, and that the governance question is who is authorised to declare it crossed. The reading is right about the question and wrong about the operator, and the correction is worth making because it changes which moves are still available.
The corpus does not put irreversibility at $K$. ch03 puts it at $F$: “Below threshold, nothing irreversible occurs; the system relaxes. At or above threshold, the fold becomes geometrically inevitable.” Week 6 of the curriculum says it in the second person — a public log entry “is the first irreversible step, which is exactly what the $F$ operator is.”
So the assignment is: $F$ is the commitment. $K$ is the last place a decision can still be taken. Once $\kappa \ge \kappa^*$ the fold follows; before that, the state relaxes. Three things follow for governance, and they follow from the shape rather than from any theorem about the atmosphere:
None of this is derivable from T1 or T2, and the paper does not claim it is. It is the analogy the arc was already running on without stating it — and the failure to state it is how a theorem about operator order came to be read as a licence to fire.
WP67 documents a distributed ice-nucleation deployment that has been running since roughly the middle of the twentieth century: microplastics nucleate ice, atmospheric weathering makes them better at it, and they are in cloud water at Mt Fuji, in Arctic snow at up to 14 400 particles L$^{-1}$, and in Antarctic snow 6 000 km from any source. No authorisation, no dose control, no attribution, no off switch.
Score it on both readings and they agree from opposite ends. Physically it is layer-local, so by K1 it is not a gate, so it does not buy the order-dependence that was the reason to want one. Politically the fold fired — continuously, for fifty years — and the gate never fired at all, because no one ever declared that an experiment was being run. It is an irreversible commitment that purchases none of the leverage the framework was after.
Four conditions. The first two are theorems about the box model; the last two are conditions on the use.
A candidate that cannot answer all four is in $O$. The screen costs one line to run, and six papers to have.