Reading a Claim Until It Breaks

Extraordinary claims usually arrive as a single headline: a device does this, a corpus shows that, a mechanism explains the other. The headline is never the thing to test. It is a compression of many smaller propositions, most of which are load-bearing and unstated. The discipline is to decompress it.

Anomaly decomposition takes a complex claim and breaks it into atomic dependencies: the individual propositions that must each be true for the headline to hold. Each is assigned an evidentiary weight and a failure mode. The dependencies are then arranged as a graph, so that the claim's actual structure becomes visible, including the joints where it can be snapped.

The useful question is rarely is the headline true. It is which propositions is the headline standing on, and what happens when you remove each one. Counterfactual removal is the sharpest tool here. Take out one dependency and ask whether the claim degrades gracefully or collapses. A claim that collapses when a single unverified proposition is removed was never as strong as its headline suggested.

What survives decomposition is often not the original claim but a residue: a narrower, better-specified statement that the evidence actually supports. Sometimes that residue is genuinely interesting and becomes a wager worth instrumenting. Sometimes nothing survives, and the exercise has still paid for itself, because it has converted an argument that could run forever into a structure that terminated.

The method is deliberately hostile and deliberately fair. It does not decide in advance that a claim is false. It decides that a claim is only as strong as its weakest load-bearing proposition, and it goes looking for that proposition first. The claims that pass are the ones where the search failed. Those are rare, and they are the ones worth the Yard's time.

Dismantle the headline; keep whatever the evidence was actually holding up.