↩ ExploRational · A Catalogue of Divergent Rationality
explorational · a divergent move
Positions cannot negotiate; cruxes can. What is the smallest claim that, if you came to believe it false, would actually change your mind — and where does yours meet theirs?
Watch a disagreement about remote work run for an hour. Each side restates its conclusion — "collaboration will die," "talent will walk" — in fresh words, at rising volume, and nothing moves. Nothing can move: a position is the output of someone's reasoning, and an argument aimed at an output has no surface to grip. The claims underneath are another matter. Beneath "stay office-first" sits something like "juniors cannot be mentored over video" — a claim about the world, with evidence for and against, and a way of being wrong.
The double-crux format, from contemporary rationalist practice (CFAR), turns that observation into a procedure. Each side descends from its conclusion and lists the load-bearing claims beneath it, then marks the special ones: a crux is a claim whose loss would flip the holder's mind; disappointment is insufficient. Lay the two lists side by side. Where a marked claim appears in both columns — a double crux — the standoff compresses to a single question both parties have already certified as decisive, and inquiry finally has an address. The method carries a confession in its pocket, too: if you descend and find nothing to mark — no claim you could not keep your position without — you have found a commitment, not a belief. Also worth knowing.
the signature instrument
Staff both columns yourself — you usually can — or sit a partner at one. Each claim carries two toggles only its holder may set: I believe this, and the crux toggle — losing this would change my position. Everything below the columns is computed from the toggles and the text: counts, matches, gates. The bench certifies overlap; it never grades your reasoning.
A remote-work disagreement
two columns · toggles are the ground truth · matches & gates computed live
The displayed values are structural — counts of toggles and lexical matches. The bench certifies overlap, not sincerity.
descend from the conclusion
A position is a verdict: "go remote-first." It compresses dozens of beliefs — about mentoring, hiring, focus, culture, rent — into three words, and the compression is the problem, because two people can trade verdicts all afternoon without ever touching the beliefs where they actually differ. Descending opens the commitment dimension by asking: what would have to be false for me to abandon this? Each answer is a rung, and each rung is smaller, more factual, and more testable than the position above it.
The crux toggle keeps the descent honest. Most claims under a position are decoration — believed, relevant, and dispensable: lose one and the position survives on the rest. A crux is structural. Marking it is an act of self-knowledge before it is a move in any debate: you are declaring, where you can be held to it, which part of your scaffolding actually bears weight. The bench counts those declarations and matches them; only you can make them true.
what to try
The bench finds one double crux in the exemplar — "Juniors can be mentored well over video" — marked crux by both columns, believed by A, disbelieved by B. Flip B's believe toggle: the card's reading changes from a live front to descend again, because a hinge the parties already agree on cannot be what separates them. Then staff the columns with a disagreement you are actually in, and find the claim neither of you would survive losing.
The rough candidate — "Async writing can replace most meetings" ≈ "Most meetings can be replaced by async writing" — shares 4/5 content tokens, so the bench suggests it but refuses to certify it. Edit one wording to match the other and watch it graduate into a second double crux, arriving UNTESTED. The gate will not mark it ACTIONABLE until an observation is named in the test field — and notice the gate checks presence, not quality. Judging the test is your job.
Untick both crux toggles in one column. The mirror appears: this column holds a commitment, not a belief — also worth knowing. No judgment attaches — commitments are legitimate things to have. But inquiry cannot reach them, and it is better to learn that in one minute at the bench than one hour into the argument.
the meeting point
Inquiry is expensive — trials, evidence, calendar time — so it should be spent where a result would actually move somebody. That is the entire logic of the overlap requirement. A claim that is a crux for you alone is a private hinge: settle it and one mind moves. A claim marked by both is the only place where one experiment can move two. The double crux is not a compromise position, and it is not common ground. It is an address — the coordinates where the disagreement actually lives — and the shared claims the bench lists without a hinge show how much of what you were arguing about was never load-bearing for anyone.
When the columns share nothing, the bench says so, and each honest continuation is a move of its own. Descend another layer — cruxes often sit one rung below where you first stopped. Split the question — "should we go remote" may be two questions wearing one sentence, each with cruxes of its own. Or notice that the columns keep failing to touch because the difference is not factual at all: you weigh the same facts differently. Naming a values difference is not defeat; it ends a fact-fight that no amount of evidence was ever going to win.
the honest bench
The matching is lexical, and it says so. Exact means exact after normalization — lowercased, punctuation stripped, whitespace collapsed — and the near-match suggester is plain content-token overlap, labeled rough: a prompt to align wording, never a verdict of sameness. The bench cannot tell whether two sentences mean the same claim; only the two of you can, which is why certification runs through your own edits and toggles rather than through the matcher.
Nor can it detect a crux named in bad faith — marked "would change my mind" by a mind with no intention of changing. The test-design gate is the working defence: sham cruxes resist naming a settling observation, and at the bench that resistance is visible to both parties as a required field that stays empty.
the move ↔ the bench
| The move | The bench |
|---|---|
| the position | the conclusion you defend — terminal, so arguments aimed at it have nothing to grip. |
| a claim | a load-bearing belief beneath the position, small enough to be wrong about. |
| the crux toggle | "losing this would actually move me" — the holder's own certification, which no matcher can supply. |
| the double crux | the claim where your hinge is also theirs — the address where the disagreement actually lives. |
| the test | the observation both sides accept in advance as settling it; the gate that makes a crux actionable. |
| the no-crux banner | a commitment recognizing itself — also worth knowing. |
how this opening fails
risk · token opposition
The catalogue's named risk for this move. The bench records toggles; it cannot audit them. A player can mark "losing this would change my position" over a claim that privately would not — token opposition in its purest form: the ritual of openness performed over a closed position. The test-design gate is the practical counter, because a sham crux resists its own test. Every proposed observation gets amended, deferred, or hedged, since naming one would arm it. Watch for the crux whose test field stays empty through three drafts — the field is measuring something, just not what it asks for.
risk · token opposition, by descent
The descent can itself become the evasion. There is always another layer beneath the current one, always a finer phrasing of the hinge, and a player who prefers not to risk their position can descend forever — impeccably rational, never testable. The format's teeth are in the gate: a double crux is supposed to end in an observation with a date on it. If three sessions produce ever-finer cruxes and no test anyone has agreed to run, the bench is being used as a fencing salle — the moves rehearsed, the touch never scored.
Find the claim you could not keep your position without — then find out if it is also theirs.
can you use it?
RECOGNITION — Which is finding the crux? A: listing every assumption under a disagreement. B: naming the smallest claim that, if it flipped, would change your mind — and asking the other side for theirs. C: restating your strongest argument.
B. The crux is the single load-bearing belief whose reversal moves you; the move trades cruxes to find where inquiry should go. A maps everything; C doesn't locate the pivot.
THE NEAREST NEIGHBOR — Surfacing the assumptions also digs beneath a claim. What separates finding the crux from it?
Selectivity. Surfacing maps the whole pyramid of premises; finding the crux isolates the one premise that actually decides the question for you, so a shared search can aim at it rather than at all of them.
PRODUCTION — You disagree with a colleague about hiring a generalist vs. a specialist. Name your crux — the one fact that, if true, would flip you — and phrase the question you'd put to them.
A version: 'My crux: if the role's needs will change every quarter, I want the generalist. What would have to be true about the role for you to prefer the other?' Yours works if flipping the named fact actually changes your position — if nothing would, you haven't found it.