Software — Polis (Computational Democracy)
Classification: consensus / opinion-clustering platform — mass open-ended participation with statistical opinion mapping; not a deliberation-room or selection tool. Updated: 2026-10-05. Evidence labels per RESEARCH-METHOD.md. Fit: unresolved — candidate for later CA-R3 assessment.
Identity
- Project: Polis; stewarded by the Computational Democracy Project (compdemocracy).
- Repository: https://github.com/compdemocracy/polis — monorepo; created 2016-11; active (pushed 2026-10-05); ~1,214 stars; Licence AGPL-3.0 (SRC-0091).
- Hosted: pol.is; self-hosting documented (Docker; third-party services needed for e.g. translation/spam filtering) (SRC-0091).
- Methods paper: Polis: Scaling Deliberation by Mapping High Dimensional Opinion Spaces (SRC-0092).
Core purpose
- Observed: gather open-ended feedback from very large groups; participants submit short statements and vote agree/disagree/pass on others’ statements presented one at a time; a real-time analysis maps “opinion groups” and surfaces statements with cross-group support (SRC-0092, SRC-0093).
Selection / sortition
- No capability found in reviewed documentation. No random selection of participants — participation is open/invited at scale.
Participation and deliberation
- Observed: this is mass opinion discovery, not small-group deliberation: no threaded discussion, no facilitated dialogue; structure is statements + votes + clustering (SRC-0092, SRC-0093).
- Method detail (SRC-0092): vote matrix (+1 agree / −1 disagree / 0 pass); participants voting on fewer than seven comments excluded; PCA dimensionality reduction (power iteration) then clustering into opinion groups; reports distinguish consensus statements (≥60% one-way overall) from divisive statements; “areas of uncertainty” (>30% pass) surfaced (SRC-0094).
- Analysis: algorithmic outputs describe distributions of expressed opinion, not decisions or “consensus” in the deliberative sense — outputs must be presented as such to avoid overstating legitimacy.
Identity and privacy
- Observed: participants can be anonymous; exports refer to participants by integer IDs unique to a conversation (SRC-0093, SRC-0094). Optional authentication/configurations exist; no eligibility-verification machinery found.
API / export / interoperability
- Observed: public read API endpoints (e.g.
/api/v3/reportExport/.../summary.csv,comment-groups.csv); conversation exports produce Summary, Stats History, Participants Votes and Comments tables (SRC-0093, SRC-0094). Strongest documented API/export posture among the reviewed deliberation tools.
Deployment evidence
- Hosted service pol.is used for large-scale public conversations (e.g. national-level exercises in Taiwan’s vTaiwan context and elsewhere, widely documented); 40+ country usage claims belong to the maintainers (SRC-0091, SRC-0092).
Limitations / open questions
- Self-organising input cannot distinguish invitation-sample participants from the general public — selection remains external.
- LLM-era derivatives/forks exist (e.g.
polis-communityefforts); canonical project’s exact maintenance model should be pinned at CA-R3.
Sources
- SRC-0091 (repo), SRC-0092 (methods paper), SRC-0093 (documentation overview), SRC-0094 (report/API evidence).
CA-R3 pinned assessment
- Pinned source/ref:
compdemocracy/polis@80193c65ff6b929a99e069ace8b00f43b52dcb80(SRC-0149); no GitHub latest-release artefact is published. - Licence: AGPL-3.0 with repository-stated additional permission under section 7.
- Maintenance/deployment: active Docker-based stack; production hardening/scaling remains operator work.
- Relevant CA-REQ coverage: strong for CA-REQ-096 mass input and opinion-group preservation; partial dissent visibility; no authority, selection or binding-decision semantics.
- API/integration seam: public API and CSV/report exports.
- Identity/privacy: conversation-scoped participant IDs; optional auth; not eligibility verification.
- Operational burden: medium-high because web clients, server, maths/analytics and storage are separate concerns.
- Material risk: analytics can be rhetorically mistaken for an assembly decision.
- Provisional classification: integrate for mass-input/opinion mapping. Polis analytics are not an assembly decision.