Double Crux

Most disagreements are terrible. Two people bash their conclusions together, get nowhere, and either escalate into hostility or retreat into "agree to disagree" — which preserves the friendship at the expense of ever actually learning anything. Double crux is CFAR's attempt to build a better format: a disagreement protocol that's collaborative rather than adversarial, designed to make both parties want to find the point where their reasoning diverges.¹

The Mechanism

You believe A. I believe not-A. Instead of arguing directly about A, we each search inward for the crux of our belief — some underlying proposition B that is both necessary for our position and likely to be where we actually disagree. A double crux is a B where:

• We disagree about B as well (I think B, you think not-B)
• B is crucial for my belief in A — if B turned out to be false, I'd abandon A
• Not-B is crucial for your belief in not-A — if B turned out to be true, you'd abandon not-A

Finding B is the victory condition, not settling B. Once you've found a double crux, you've already made enormous progress, because you now have a shared theory about the causal structure of the universe — you agree on what would change each other's minds. That's a collaboration, not a fight.¹

For example: you think middle school students should wear uniforms (A). I think they shouldn't (not-A). After searching for cruxes, we might converge on B: "uniforms significantly reduce bullying based on economic class distinctions." I believe this; you don't. Now we have a concrete, empirically investigable question instead of a vague debate about school policy. Even better, we can iterate — maybe B isn't quite right either, and there's a C underneath it that's even more concrete.

Prerequisites That Sound Easy and Aren't

The technique requires four things that most people endorse in theory and violate constantly in practice:

Epistemic humility. The willingness to think "I might be the one who's wrong here." This sounds obvious but goes against deeply wired tribal instincts. Treating your own beliefs as objects you can examine and provisionally set aside, rather than extensions of your identity, is a skill that takes practice.

Good faith. An assumption that people believe things for causal reasons — that if you'd been exposed to the same information and experiences, you'd hold approximately the same beliefs. This is the opposite of the common debate move of attributing your opponent's position to stupidity, bias, or malice.

Confidence in objective truth. For almost any well-defined question, there really is a fact of the matter. It might be impractically hard to discover, but it exists. Without this conviction, there's no reason to keep digging — you can just retreat into relativism.

Genuine curiosity. You have to actually want to understand what the other person sees that you don't. Not as a strategy to find their weaknesses, but because the universe is complex and you probably only see part of it.

The Weak Version

The strong version of double crux requires both parties to understand the protocol and consent to it. But you can also use it unilaterally — the "weak version" — by embodying the question "What do you see that I don't?" This means approaching from genuine humility, drawing out your interlocutor's belief structure for its own sake rather than to undermine it, and modelling the vulnerability you'd like them to show.

The key insight is that people can smell fake good faith. If you're running the double crux algorithm as a manipulation technique — "let me get them to expose their reasoning so I can demolish it" — it won't work. The technique requires, and perhaps its most important function is to cultivate, the actual stance of collaborative truth-seeking rather than competitive debate.

Why Disagreements Dissolve

One of the most underrated features of double crux is that disagreements often simply disappear when you operationalise them carefully. "Should we invest more in research?" is an intractable disagreement. "Of our next $10,000, how much should go to research vs. advertising?" suddenly produces a lot of agreement, because the abstract principle was masking a shared practical understanding. Similarly, many political and philosophical disagreements turn out to be about the meaning of words rather than the state of the world. This isn't a failure of double crux — it's one of its best outcomes.¹

This connects to a broader pattern in rationalist epistemology. The maps-and-territories framework reminds us that when two maps disagree, the disagreement is often in the map-making rather than in what the territory actually looks like. Double crux is essentially a protocol for comparing map-making procedures rather than maps, which is why it sometimes reveals that the maps were never as different as they appeared.

Limitations

Double crux works best between people who share enough epistemic framework to have meaningful cruxes. It works less well when the disagreement is fundamentally about values rather than facts — "we should prioritise equality over freedom" doesn't have a clear B that both parties would accept as decisive. It also struggles when one party's actual crux is "I trust my group's authority on this" rather than any object-level proposition, because authority-deference doesn't decompose into falsifiable sub-beliefs.

And of course, most environments punish the vulnerability that double crux requires. Admitting what would change your mind is dangerous in any setting where your interlocutor might weaponise that admission. The protocol works in high-trust, small-group settings. Scaling it to public discourse or institutional decision-making remains an unsolved problem — perhaps unsolvable, given that adversarial dynamics are often adaptive rather than pathological.