Facilitator guide

Good fit:

• The question is open — "where should we go" rather than "A or B".
• You want the distribution of views: how many camps, what each cares about, where the common ground is.
• Participants share basic trust (communities, classes, neighborhoods, internal teams, workshops).
• Pre-meeting input: lay out the disagreements before the meeting so the meeting goes straight to them.

Poor fit:

• Settings that need named speech or formal votes (everything here is anonymous).
• Adversarial public controversies with ballot-stuffing incentives (use official pol.is or identity-verified tools).
• Groups of two or three (just talk; clustering starts at 4 participants).

Frame the topic as an open question, e.g. "What should our neighborhood invest in over the next two years?" Avoid yes/no questions.

Seed statements are the 8–15 sentences that exist when the conversation opens, so early participants have something to vote on. Guidelines:

• One idea per statement — "add bike lanes and improve streetlights" should be two statements.
• Agreeable or disagreeable — write positions, not questions or facts.
• Cover the spectrum — deliberately include views you disagree with but that exist in your community; the map should show the whole landscape.
• Include a few likely-consensus statements — consensus needs to be made visible too; these will surface in the common-ground list.
• Plain, concrete, under 280 characters. "Improve communication" is too vague; "Major decisions should be announced a week in advance" is votable.

• New statements publish immediately: keep it on inside a trusted circle — it keeps the conversation alive. Turn it off for public links or sensitive topics and approve each statement yourself (works fine from a phone).
• Participants can add statements: this is the soul of the Polis method — keep it on unless you are purely testing fixed options.
• Public data export: recommended for academic and civic-tech settings so anyone can verify your conclusions; usually off for internal discussions.

• From 4 participants clustering begins (each must vote past the threshold: all statements when there are fewer than 7, otherwise 7).
• Above ~20 the map stabilizes; above ~50 the representative and consensus lists become reliable.
• Live workshops: 10–20 minutes of voting (20–40 statements). Async online: run 3–7 days with one mid-way reminder.
• The system is designed for hundreds to low thousands of participants and hundreds of statements.

1. 3-minute opening: explain anonymity (script below), demo one vote, stress that "Pass" is a legitimate choice.
2. 10–20 minutes of voting: project the participation link as a QR code (any QR generator works). Encourage people to add their own statements after voting — new statements get shown to others first.
3. Keep the host page open: approve new statements (if moderation is on) and unpublish obvious duplicates.
4. Read the map together, 10 minutes: project the results page. Common ground first ("look what we all agree on"), then each group's distinctive statements ("here is where we actually differ"). Don't rush to debate; let the structure sink in.
5. Bridge to next steps: turn consensus statements into an action list; turn the deepest divides into the agenda for the next (deliberative) session.

"This tool is anonymous: no registration, no name or email. The system keeps one random code in your browser so the same person isn't counted twice. Your votes become statistics only; each dot on the map shows that someone is there, not who they are. Statements you submit are shown publicly but never attributed."

Imagine laying out each person's full voting record in a row. The more similarly two people voted, the closer they are placed; flattened into a two-dimensional map, people who voted alike sit together, and the system circles the clusters into groups (it tries 2 to 5 groups and keeps the cleanest split).

A group's "distinctive statements" are the ones where that group's voting differs most from everyone else — with a statistical test guarding against flukes. "Common ground" is the statements every group leans toward agreeing with; if even one group doesn't, it stays off the list.

The method is reimplemented from the published Polis literature and validated against official open conversation data (our clustering closely matches the official runs). For the math: algorithm notes, validation report.

• Close voting from the host page when done (the results stay viewable) and download both CSVs for the record.
• Summary order: participation numbers → consensus list → the groups and their distinctive views → next steps. Lead with consensus — it is this method's most valuable output.
• Turn consensus statements into proposals or action items; carry the deepest divides into the next deliberative session.

• The anonymous code lives in the browser: a new device means a new identity. It prevents accidents, not determined ballot-stuffing.
• The map shows the structure of voting behavior — it is not a representative poll; participants are not the population.
• Groups are an algorithmic summary, not labels. Read the map as "how this group voted", never "who these people are".