Practice · reopen

How to talk to someone inside a closed claim

You cannot argue a closed claim into the open, because the claim has already decided what your argument is: an attack, a lie, a sign you are one of them. What you can do is reopen one of the cut paths — by asking rather than asserting, and by naming the cut rather than the answer. This page collects the site’s counter-moves in one place.

What you are trying to do

Not win. A closed claim is protected against losing; a conversation aimed at winning confirms the protection. The aim is smaller and more useful: to get one of the four error-correction paths working again, so that evidence can enter the conversation at all. Once a path is open, the ordinary argument about what is true can begin. Until then it cannot, and pretending otherwise is why these conversations go the way they go.

The person you are talking to is not stupid, and telling them so — in words or in tone — hands the claim its best evidence. The candidate mechanisms behind closure work on intelligent people, and several of them run on the feeling of being looked down on. Assume you are talking to someone who would revise if a path were open, and work on the path.

Three rules

1. Grant the emotion, decline the warrant. The feeling is usually real and often justified: distrust of institutions that have failed, contempt that has actually been directed at them, a loss that actually hurt. Grant it fully. Then separate it from the inference: “That can all be true, and the claim can still be false. They are independent.” Denying the feeling reclassifies you; granting it leaves only the inference to argue.
2. Ask; do not assert. An assertion can be recoded as an attack from a disqualified source. A question is harder to recode, because the answer has to come from the speaker. The four questions below are designed to be asked out loud, in a tone of genuine curiosity, one at a time.
3. One claim, until it resolves. Closure travels on volume. When the first allegation fails and a second appears, do not follow it. “We can do that one next. This one first: did it hold up?” The flood only works if you drink all of it.

The four questions, said out loud

The site’s one test, in conversational form. The check page runs the same four questions on a claim; here they are phrased for a person.

1. What is actually being claimed? “Help me pin it down. In your telling, what exactly happened — who did what, and when?”
   State it in a form that evidence could confirm or disconfirm. If the honest answer is the closed one — it cannot be pinned down: the wording shifts, or no observation could count against it — you have found the cut. Name it gently, and stay there.
2. What is treated as already settled? “Can I ask about the part we’re both taking for granted? How do we know that bit?”
   Look for the premise smuggled in as background rather than argued for. If the honest answer is the closed one — a contested premise is presented as background that only a fool or an enemy would question — you have found the cut. Name it gently, and stay there.
3. Who is allowed to check it? “Is there anyone you would trust to check this — anyone at all, even someone you don’t like?”
   A claim with no acceptable referee has no working error-correction path. If the honest answer is the closed one — every referee who could return an adverse finding is disqualified in advance or on arrival — you have found the cut. Name it gently, and stay there.
4. What result would change the speaker’s mind? “Before we look at anything: what would convince you otherwise?”
   If the answer is “none,” the dispute has been closed, not won. If the honest answer is the closed one — no outcome is named that would count against the claim; wins and losses both confirm it — you have found the cut. Name it gently, and stay there.

The Popper question

The fourth question is the one that matters most, and it should be asked before any evidence is presented, not after. Evidence presented first arrives as an attack; the same evidence presented after the speaker has named what would count arrives as an answer to their own question.

“Before we look: what would convince you otherwise? If the answer is ‘nothing,’ the belief isn’t about evidence.”

Say the first sentence and stop. Do not say the second sentence to the person; it is for you. If they name something, you now have a shared test and can go and look together. If they cannot, the conversation is not about evidence, and no amount of it will help. That is not a defeat; it is the information you came for.

The counter-move for each construction

The counter-move is the one the closing script is written to avoid: answer the question that was skipped. Each device in the catalog names the question it makes unaskable and a way to ask it anyway. They are collected here from the dossiers; the device pages show the specimens.

Closure devices

D1 Persecution transfer

Forbidden question: Did he actually do the thing charged?

Grant the feeling, decline the warrant: “Your sense of being looked down on can be fully legitimate, and he can still have done it. Those are independent. Which evidence would you accept?”

D2 Referee removal

Forbidden question: Is this particular report accurate?

De-categorize: “Pick this one story. What did it get wrong, specifically? If nothing, the label is doing the work the evidence was supposed to do.”

D3 Evidence laundering

Forbidden question: Who says, and on what basis?

Run provenance out loud: “Name one of the many people. If the trail ends at the speaker, we’ve learned what we needed.”

D4 Savior monopoly

Forbidden question: Compared with which actual alternatives, on which records?

Re-pluralize: “Suppose he’s right that it’s broken. What are the other three ways to fix it, and why are they worse? If that question is unwelcome, note why.”

D5 Heads-I-win priors

Forbidden question: What evidence would show the loss was legitimate?

The Popper question, gently: “Before we look: what result would convince you it was legitimate? If the answer is ‘none,’ the belief isn’t about evidence.”

D6 Retroactive frame control

Forbidden question: Did he mean it — and does meaning even matter once uptake occurred?

Freeze the ledger at uptake: “Millions heard it straight. Which correction reached them?”

D7 Moral recategorization

Forbidden question: Are these citizens with contestable views — and what, exactly, did each do?

Re-personize: “Name one specific person covered by that word and the specific act that earns it. Categories don’t commit acts.”

D8 Voice merger

Forbidden question: Is he in fact serving your interests — and how would you tell?

Reopen the audit: “If he’s your voice, you’re the boss. Bosses check the work. What would firing-quality performance look like?”

Force multipliers

M1 Nominal verdict

Forbidden question: Is the claim in the name true, and what follows from their actual arguments?

Force the predicate: “Crooked how, specifically, and what’s the evidence? A name is not a finding.”

M2 Apocalyptic license

Forbidden question: Are the stakes actually that — by what measure?

Demand the threat model: “Losing means no country — via what causal chain? Name the mechanism and we can check it; no mechanism, no license.”

M3 Zone flooding

Forbidden question: Can we settle one claim before the next? (Made moot, not asked)

Refuse the buffet: “One claim. We stay on it until it resolves. The flood only works if you drink all of it.”

M4 Superlative saturation

Forbidden question: Compared to what baseline, in what units?

Demand units: “Greatest by which number? Here’s the series; point to it.”

What not to do

• Do not open with the disqualified referee. “The courts threw it out” is an attack from inside the plot until the referee question has been asked. Ask who they would trust first; then cite that.
• Do not lead with “that’s been debunked.” It asserts the result of a check whose legitimacy is the thing in dispute, and it is the sentence closure was built to absorb.
• Do not do it in front of an audience if you can avoid it. Several of the mechanisms are about identity; conceding in public costs more than conceding in private, and a person who cannot afford to concede will not.
• Do not argue the person to argue the claim. A voter can be respected, aggrieved, and still mistaken about a factual claim; keep the evidence and the identity in separate sentences, or the claim will merge them for you.
• Do not expect the moment. Corrections that work tend to work later, alone, and unannounced. The evidence on corrections is mixed — the backfire literature is cited on this site as a pair, a finding and its non-replication — and none of it promises a conversion on the spot.

When to stop

If the referee question and the Popper question both come back closed after being asked plainly and kindly, stop. You have learned that the claim is currently not about evidence, and continuing turns you into the persecution the claim predicted. Leave the questions behind; they are the part that works later.

Then turn the test around. Take a claim you hold, and answer the four questions about it honestly. If you cannot name a referee or a result, you have found a closed claim of your own, and the same advice applies. A conversation which never reopens a path is not going to produce a correction, whatever else it produces.

Drill the counter-moves on real sentences → Or run the four questions on a claim