What is the Native Debate System?

Our Native Debate System is a next-generation deliberation platform built directly into Society Speaks. It combines the proven principles of pol.is with powerful new features like threaded arguments, evidence linking, and optional AI-powered insights.

✨ Key Advantages:

No separate account needed (fully integrated)
Instant real-time voting with AJAX
Add detailed pro/con responses with evidence
Interactive consensus visualizations
Optional AI features (you control the costs)

How to Participate

1 Vote on Statements

Agree - Click if you support the statement
Disagree - Click if you oppose the statement
Unsure - Click if you're undecided or need more context
Votes update instantly without page reload
You can change your vote anytime

2 Add Your Own Statements

Click "Add Statement" to contribute
Keep it concise (10-500 characters)
Choose type: Claim (assertion) or Question
System checks for similar statements to avoid duplicates
You have 10 minutes to edit after posting

3 Add Detailed Responses

Click "Add Response" on any statement
Choose your position: Pro, Con, or Neutral
Write up to 2000 characters
Add evidence: citations, URLs, or file uploads
Responses can be threaded (reply to replies)

4 View Consensus Analysis

After enough participation, consensus analysis becomes available
See how different opinion groups formed
Discover statements with broad agreement
Identify "bridge" statements that unite groups
View interactive scatter plot visualization

See a live example of consensus analysis →

Sorting Options

Choose how statements are displayed:

📊 Progressive (pol.is style)

Shows you statements with fewer votes first, ensuring all get attention

⭐ Best

Highest agreement statements first (using Wilson score)

🔥 Controversial

Statements where the overall vote is split — surfaces disagreement in the statement feed. The formal "Points of Division" panel on the analysis page uses a stronger, cluster-aware test.

🕐 Recent

Newest statements first

Understanding Consensus Analysis

After enough participation (7+ users, 20+ votes), the system analyzes voting patterns to:

👥 Identify Opinion Groups

Users are clustered based on voting similarity, not demographics. The system uses PCA (Principal Component Analysis) and Agglomerative Clustering to find natural groupings.

Example: Group 1 might prioritize environmental concerns, while Group 2 prioritizes economic growth—even if both have mixed political backgrounds.

✅ Detect Consensus Statements

A statement qualifies as consensus only when the Wilson 95% score interval on overall agreement clears 70%, AND every opinion group’s own Wilson lower bound clears 60%. Raw percentages alone are never enough — 7/10 agreement is not "70% consensus" when its confidence interval starts at 40%.

These represent common ground that stands up to small-sample scrutiny.

🌉 Find Bridge Statements

Statements where every group lands on the same side — either all agreeing (≥65% mean) OR all rejecting (≤35% mean) — with low variance between groups. Shared rejection counts as common ground too.

These are especially valuable for finding solutions that work for everyone.

⚡ Highlight Divisive Statements

Statements where different groups vote differently — measured by the gap in agreement between the most- and least-agreeing groups. Significance comes from a pre-specified permutation χ² test across ALL clusters, not post-selection of extreme pairs. The gap CI uses Newcombe’s MOVER-Wilson Method 10.

These reveal where dialogue is most needed. FDR-corrected results are marked ✓; uncorrected results are shown as "suggestive only".

🔬 Statistical rigour

Wilson 95% score intervals on every reported percentage — no more "100% agreement" from 2 of 2 votes.
One-sided Fisher’s exact test for representativeness (which statements distinguish a group from everyone else).
Benjamini–Hochberg FDR correction at q=0.05 across all cluster×statement tests (or Benjamini–Yekutieli for strict peer review under dependence — configurable).
Stability metrics (Adjusted Rand Index, Jaccard overlap) computed over re-seeded clustering runs; analyses whose clusters are not reproducible are withheld.
Limitations are called out in the methodology panel of each results page: clustering is fit on the same votes used for group summaries, so all inferential quantities are exploratory rather than confirmatory.

📊 Visualize with Interactive Charts

See a 2D scatter plot where each dot is a participant, colored by opinion group.

Hover over dots to see details. Closer dots = more similar voting patterns.

👀 See It In Action

Want to see what all of this looks like? Explore our live demo discussion on climate action and its full consensus analysis.

View Discussion View Consensus Analysis

Adding Evidence

Strengthen your responses with supporting evidence. You can add three types:

📖 Citations

Academic papers, books, reports

🔗 URLs

Links to articles, studies, or websites

📎 File Uploads

PDFs, images, or documents

💡 Evidence can be marked as verified, pending, or disputed by moderators.

Moderation & Safety

We maintain high-quality discussions through:

User Flagging: Report spam, offensive content, off-topic posts, or duplicates
Edit Window: 10-minute grace period to fix typos or clarify, then content becomes immutable
Moderation Queue: Discussion owners can review flagged content
Rate Limits: 30 votes per minute, 10 statements per hour
Audit Logs: All moderation actions are tracked for transparency

Optional AI Features

Unlock advanced features with your own AI API key (OpenAI or Anthropic):

🤖 AI Discussion Summaries

Generate human-readable summaries of consensus analysis results, highlighting key agreements and disagreements.

🏷️ Automatic Cluster Labels

AI generates descriptive names for opinion groups based on their shared beliefs (e.g., "Environmental Advocates" vs "Economic Pragmatists").

🔍 Semantic Deduplication

Prevents semantically similar statements (not just exact duplicates), helping users find and vote on existing statements instead.

💰 Cost: ~$0.01-0.05 per analysis. You control when AI features are used, and your API keys are encrypted before storage.

Manage API Keys →

Native System vs Pol.is

Both systems are excellent, but they serve different needs:

Feature	Pol.is Embed	Native System
Account Required	Separate pol.is account	✓ Society Speaks only
Real-time Voting	Page reload required	✓ Instant AJAX updates
Threaded Responses	No	✓ Pro/con threading
Evidence Linking	No	✓ Citations + files
Edit Window	Immutable	✓ 10-minute grace
AI Summaries	No	✓ Optional (user-key)
Mobile UX	Good	✓ Tailwind optimized
Best For	Quick large-scale polls	Deep deliberative discussions

Native Debate System