Verified disarmament
A Socratic walk-through of verified disarmament — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can a weapons treaty be enforced by counting things that are not the weapons themselves?
A treaty limits the number of warheads a state may hold. Warheads are small, easily hidden, and among the most closely guarded objects any state possesses. No inspection regime is going to walk through every building in a country and count them.
So the treaties count something else. They count missile tubes, silos, bombers, launchers — and they count them from satellites, at declared sites, with rules about what must be left visible. The obligation is stated in warheads and the verification is performed on hardware nobody is trying to prohibit. That looks like a substitution of the measurable for the meaningful, which is usually a mistake. Here it works, and it is worth understanding why.
Reasoning it through
REASONING #Start by asking what verification actually needs to achieve. The naive goal is to know how many warheads exist. That is unattainable, and pursuing it produces a regime that either fails or demands intrusion no state will accept.
The achievable goal is different: make cheating detectable at the scale that would matter. Nobody needs to catch a single concealed warhead. What matters is detecting a breakout large enough to change the strategic balance — and a breakout of that size is not a hiding problem, it is a deployment problem.
That reframing does the work. Ask what a state must do to convert hidden warheads into usable strategic capability. It must have delivery systems: missiles, and something to launch them from. Those are large, they require distinctive infrastructure, they must be tested, and they cannot be hidden from overhead observation at scale. A silo is a hole in the ground with a particular signature. A submarine has a fixed number of tubes. A bomber sits on a runway.
So the deliverable capability is bounded by the delivery systems, and the delivery systems are exactly the things that cannot be concealed. The treaty counts the observable term in the product, because the unobservable term is useless without it.
The second device closes the remaining gap. Since a single launcher might carry several warheads, treaties use attribution rules: each type of missile is counted as carrying an agreed number, whether or not that is what it holds today. This looks like a fiction and is in fact the cleverest part. It converts an unverifiable question — how many are on that missile right now — into a verifiable one — how many missiles of that type exist. The number attributed can be negotiated, and if it errs it errs conservatively.
The third device is designed observability. Treaties do not merely permit inspection; they require states to make the relevant facts visible. Bombers must be parked so overhead sensors can see their configuration. Submarine tubes may have to be visibly disabled. Destroyed items must be destroyed in specified ways leaving verifiable remains, not simply reported gone. The regime obliges each side to produce evidence the other can check.
Put together, these give the structure. Do not verify the prohibited object; verify a necessary complement of it that cannot be hidden, then fix the ratio by agreement, then require the complement to be displayed.
One more element is needed or the rest fails: both sides want to be verified. A state that has genuinely reduced gains nothing from a reduction the other does not believe. The regime is not purely adversarial — it is partly how each side makes its own compliance credible, which is why states accept intrusions they could refuse.
The analogy
THE ANALOGY #Think of a licensing scheme for commercial fishing that cannot possibly count fish in the sea or in every hold.
It counts boats, licences, net dimensions and days at sea. None of those is a fish. But catch is bounded by capacity, capacity is embodied in equipment that is large and registered, and the equipment must be used in the open to be useful. A vessel could conceal some of its catch; it cannot conceal being a vessel. And the regime does not merely watch — it requires transponders on, logbooks kept, landings made at designated ports.
A fishing fleet's capacity translates fairly directly into catch, whereas the relationship between launchers and warheads is set by an agreed convention rather than by physics — so the arms-control version depends on a negotiated number in a way the fishing analogy does not, and that number is a standing subject of dispute.
Clarifying the model
THE MODEL #Attribution rules are a negotiated fiction, and the negotiation is where the real argument is. Counting a missile as carrying an agreed number is only as good as the agreement. Change the number and the same hardware yields a different treaty-limited total; a system's attribution is therefore fought over, and a state that can carry more than it is attributed has an upload capacity that sits legally inside the treaty. This is a genuine limitation, not a detail, and it is the standard criticism of launcher-counting regimes.
The approach works for delivery-limited weapons and works badly elsewhere. The whole argument rests on the prohibited item being useless without a large, observable complement. Chemical and biological agents do not have that property — production can be small, dual-use and concealable, and the delivery requirement is far weaker. That is why verification in those domains relies on different and less satisfactory instruments: declarations, materials accounting, challenge inspections, and monitoring of precursors and equipment. Anyone generalising the launcher-counting success to all arms control is overstating it.
Detection is probabilistic, and the deterrent is the risk rather than the certainty. No regime guarantees that violation is caught. What it does is raise the probability of detection for violations large enough to matter, and pair it with consequences. The design question is always the same trade: how much intrusion will be accepted against how much confidence is bought.
National technical means do much of the work. Much verification is done by each party's own satellites rather than a neutral inspectorate, with treaties committing both sides not to interfere with them. That is efficient between parties who both have such capabilities and awkward where they do not.
Both sides of the political dispute are interested parties. Whether a regime has been complied with is argued by governments and by analysts funded on both sides. I am describing the structure of verification, not adjudicating any compliance claim.
The falsification test. If the mechanism is bounding capability through an unhideable complement, then a regime should hold where the prohibited item requires large observable delivery infrastructure, and should fail where it does not. If launcher-counting regimes and agent-based ones had proved equally verifiable, the account here would be wrong and verification success would depend on something else — political will, or inspection access as such.
A picture of it
THE PICTURE #How to readThis is a requirements chart repurposed to compare what each verification means can actually establish, so read the boxes as things a regime needs and the elements as the instruments available. The pattern to notice is the single traces link, which marks the one requirement no instrument satisfies — counting the warheads themselves. Every other link is a satisfies, and together they reach the goal by a different route: bound the capability, fix the ratio by convention, require display. The risk fields are argued rather than measured.
What became clearer
WHAT CLEARED #Verification does not work by observing the prohibited thing. It works by finding a necessary complement that cannot be hidden — the launcher, not the warhead — bounding capability through that, and then converting the remaining unobservable ratio into a negotiated number. Add an obligation on each party to display the counted items and you have a regime that gives useful confidence without anyone opening every door. It is a deliberate substitution of a checkable proxy for an uncheckable fact, and it is honest about being one, which is precisely why the attribution number is where the arguing happens.
Where to go next
ONWARD #- Why upload capacity sits legally inside launcher-counting treaties, and what that does to their value.
- How chemical and biological verification differs, and why it is harder.
- Why states accept intrusive inspection at all, and what they buy with it.