Spectrum Street Epistemology: the method

Street Epistemology (SE) is a way of talking with someone about how they know what they believe, rather than arguing about whether it is true. The spectrum variant, popularised by Peter Boghossian, adds a physical line so a whole room can play at once and so that movement is visible.

The setup

Seven lines on the ground, left to right:

Position	Label
-3	Strongly disagree
-2	Disagree
-1	Slightly disagree
0	Neutral
+1	Slightly agree
+2	Agree
+3	Strongly agree

The facilitator reads a claim. A claim is one declarative sentence, not a question: "Trans women should be able to compete in women's sports." Everyone starts at Neutral. Then a count from five, and on "move" everyone walks to a line.

The rules

• You can't stand between the lines. Pick one of the seven. "Somewhere between agree and slightly agree" is not an answer; forcing a choice is what makes the reasons come out.
• You can't move until the moderator says move. Everyone chooses at the same moment so nobody's choice is shaped by watching where others go.
• After that, you can move at any time. If something you hear changes your mind, move. Everyone sees it, and the moderator will often notice and ask what moved you. When a good point lands, several people may move at once; the moderator can ask whether that point is what did it.

The app enforces these. Positions are integers, and a first choice is visible only to its owner until the facilitator reveals. After the reveal, dragging your avatar to another line (or tapping one) moves you at once, in public, and the move goes into the transcript.

The questions

The facilitator's job is not to argue. It is to find out whether the reason a person gives is the reason that is actually holding them on that line.

• Why are you standing there? Start with the ends of the line.
• Let me make sure I have your argument. Restate it. "You think X because Y, is that fair?" Correct it until they say yes.
• Did that persuade you? After someone else speaks. If not, why not?
• If I could show you... Pick the key premise. "If I could show you evidence, to your satisfaction, that hormone therapy does not remove the advantage, would you move?"
• How far? "Would you move one line? Two? What stops you from going further?" Whatever stops them is the real reason. Talk about that.
• What do you mean by...? Any term that is doing work in the argument ("transphobic", "unfair", "ally", "woman"). Ask whether everyone in the room would define it the same way. A disagreement about definitions is often the whole disagreement.
• What would it take for you to move? The central SE question. "Nothing" is a legitimate answer, but it is worth asking whether the person holds a belief or a commitment. Beliefs respond to evidence; commitments do not.
• Do you have a question for them? Let participants interview each other.
• How do you know? "How did you come to that view? How sure are you, out of ten? If someone used the same method to reach the opposite view, would it be reliable?" SE is about the method behind a belief as much as the belief.
• Facts or values? "If you two agreed on every fact, would you still disagree?" If yes, the disagreement is about values; ask what each of them is protecting.
• Steelman. "What's the best reason for standing where they stand?" Then ask the other person whether that is their reason.

• Not that you agree, but do you understand? "Not that you agree, but do you understand what Ben is saying?" Understanding is a separate thing from agreeing, and people often find they had neither.

• Let them think. When someone is working out an answer, don't fill the silence or rescue them with a follow-up. If others jump in: "We're going to let Ada think this through." In the app, "Ada is typing…" shows under the conversation while someone types or dictates.

Guess their reason

Pick two people on opposite sides, as far apart as possible: the two ends are best. Each writes down, privately, the main reason they are standing where they are. Then:

• Ask the first person what they think the other wrote.
• Ask the other: "Is that it? Not that you agree, but did Ada understand what you're saying?"
• If it missed, give the first person another try.
• Reveal what the other wrote. Ask the first person whether they understand it now.
• Swap: the second person guesses the first person's reason, and so on.
• Ask both whether understanding the other's reason changes where they stand. Watch the room for moves.

It works best once a few reasons are out, especially when the two sides are describing each other in ways the other would not recognise. Once or twice a session is plenty. In the app, the organiser controls have Guess their reason: pick the two (it suggests the two ends), Start, then Reveal each person's reason when it's time. The two people get a box on their screen to write in; nobody else, the organiser included, sees what they wrote until it is revealed. The AI moderator runs the whole thing itself.

Narrowing the question

Most claims bundle several questions. Once people have given their own reasons, the disagreement often turns out to sit on one narrower question: a scope ("at the elite level"), a condition ("only if they transitioned before puberty"), one of the ideas in the claim, or one side's meaning of a disputed word. Name it, check with the person that you have it right, and put the narrower question to the room: "Let's narrow it. The question for now is... Stand where you are on that." People move. Then go back to the full claim and ask whether anyone stands differently on it than on the narrow one, and why. That gap is often the most revealing part of the session.

A good narrowing is one a thoughtful person could stand anywhere on. Never narrow toward your own view, or to trap someone. One at a time, and not before people have had their say. In the app, the organiser controls have Narrow and Back to the claim; the narrower question shows under the claim and rounds revealed on it are marked.

Knowing the topic helps: which considerations each side usually raises, which words they define differently, which facts are disputed. Use that to recognise what someone is reaching for and to ask a sharper question, not to correct them. The AI moderator researches the claim before the session for exactly this.

What the facilitator does not do

• Does not say which line is right.
• Does not add arguments of their own except as hypotheticals ("what if I could show you...").
• Does not let the conversation become about the facilitator. When challenged ("what's the point of this?"), answer honestly and briefly: the point is to find out whether the lines people are on are the lines they should be on, given the reasons they have.
• Does not rescue anyone from an uncomfortable question, including themselves.

Late arrivals

Someone who arrives after the reveal gets the two rules in a sentence (you can't stand between the lines; you can move any time something changes your mind), takes a place, and is asked why they are there. Their first step is taking a place, not a change of mind, and the app logs it that way.

When someone is upset or disrupts

Strong feeling is part of the game. Anger about the topic, blunt language about the claim, even "this question shouldn't be asked" are all things to explore: ask what makes the question harmful, and what would make it all right to discuss. They can stand on the line like anyone else.

Disruption is behaviour that stops the conversation: attacking people in the room, slurs, threats, shouting people down or flooding the chat, derailing again and again after being asked. The steps, in order:

• De-escalate. Slow down, acknowledge it ("this clearly matters to you"), restate the rules, invite them onto the line.
• Caution. Calm and short, about the behaviour, not the view: "Ben, talk to Ada's reasons, not about Ada."
• Mute for a few minutes. They can still listen and move on the line.
• Remove if it goes on. Remove at once only for threats, slurs aimed at someone, or posting someone's private information.

Afterwards, check briefly on whoever was targeted and go back to the claim. Don't discuss the person who was removed or ask the room to judge them. Never caution, mute or remove anyone for their view, however unpopular.

In the app the organiser's people table has Mute (5 minutes), Unmute, and × (remove, with an optional reason). A removed person is told, and can't rejoin from the same network until the organiser presses Let back in. A muted person's chat, notes and call mic are off. Anyone posting more than five lines in fifteen seconds is asked to slow down. The AI moderator follows the same steps and has to go in order; it cannot see the video, so it never mutes or removes anyone on a report alone.

Running a round in the app

With an AI moderator, Claude does steps 3 to 7 by itself; see the README.

• Create a session with the claim. Share the code or QR.
• Read the claim and the rules aloud.
• Open vote. Everyone picks on their phone. The facilitator sees who has chosen but not what.
• Reveal. Five-second countdown, then everyone's position appears.
• Give the mic (spotlight) to someone on an end. Ask why. Ask what would move them. Participants can type both on their phone as they talk, so the transcript keeps them.
• Watch for moves. Anyone can move at any time; the mover is tagged on the street and in the transcript. Ask them what moved them. If several move after one point, ask whether that point did it. To have everyone reconsider together, Open re-vote and reveal.
• Repeat. End the session and export the transcript.

Choosing a claim

Good claims are short, declarative, contested in the room, and about something people have actually thought about. Avoid claims that are questions, that contain two claims, or that are so loaded that no one will stand on one side in public. If nobody will stand on a side, that itself is worth noticing and saying aloud.