How Shotoku 2.0 works

Step 1: Collect statements

Publish a topic and 15–40 seed statements—short lines that split opinion, authored by facilitators.

Participants answer agree / disagree / pass on each statement. Like Pol.is, we rely on vote patterns—not text scraping.

Step 2: Build the opinion map

From all votes we build a participant-by-statement matrix, then run:

• PCA projects the matrix to 2D to find the axis where opinions diverge most.
• K-means clusters similar voters (2–5 clusters, tuned with silhouette score).
• Opinion map: plot participants in 2D to show who sits near which cluster.

Hundreds of voices become a handful of opinion groups and their relationships.

Step 3: Group-aware consensus (GAC)

The core is not headcounts but statements every cluster can accept—we use the same family of formulas as Pol.is.

For each statement c and cluster g, let P_agree(g,c) be the agree rate in g:

C(c) = ∏ P_v=a(g, c) ^ (1/n)
       g

Scores rise when every group agrees; divisive lines sink. Bridging statements float to the top.

Step 5: Show agreement and divides

After enough activity, we auto-build rankings of high-agreement statements and divisive fault lines.

Consensus lines become agendas for workshops or policy talks; visible divides surface the real issues to debate next.

Defenses against brigading

We combine several controls so homogeneous blocs cannot distort outcomes:

• Authentication required (Social9 account or verified email—no drive-by anonymity).
• Rate limits to block burst voting and spam.
• Time-window PCA to detect sudden shifts using recent votes.
• Adaptive down-weighting when participation looks skewed.
• Strict moderation for duplicates, personal attacks, and multi-point posts.

Case study: vTaiwan and Uber

In 2015 Taiwan used Pol.is for four weeks on Uber entry: four hostile clusters merged into two, and a two-tier rating rule for second-hand operators reached ~95% cross-group agreement and later became law.

Shotoku 2.0 aims to bring that vTaiwan-style workflow to governments, companies, and communities—not only administrations.