CA-R3 synthesis — technology and reuse assessment
CA-R3 closing synthesis. Updated: 2026-10-05. Recommendations are advisory (reuse/adapt/integrate/reference/omit/unresolved) — no implementation authority is claimed; architecture belongs to CA-A1 + operator decision.
Mature capabilities Cicada should probably not rebuild
- Stratified-selection mathematics — published, peer-reviewed, maintained (sortition-algorithms 0.12.14).
- Mass open-ended public input — Polis (and Decidim/CONSUL classes) are mature.
- Asynchronous member deliberation + decision methods — Loomio covers the method range CA-REQ-051 needs.
- Census-verified mass participation + offline ballot channels — CONSUL.
- Public participation platforms generally — Decidim/CONSUL; heavy but proven.
Credible reuse candidates
- sortition-algorithms — as the selection engine (licence boundary to architect). (
adaptof an engine; nothing needs reimplementing.) - stratification-app/groupselect-lib — secondary components in the same ecosystem (
reference/adapt).
Better as integrate
- OpenDLP — respondent→selection→replacement workflow, self-hostable or SF-operated; pre-release maturity.
- Decidim — public participation envelope + nearest accountability component; do not use for selection (“sortitions” draws proposals).
- CONSUL — mass input with census verification and the corpus’s only offline-participation machinery.
- Loomio — between-session member deliberation; multiple decision methods.
- Polis — labelled mass-input channel (opinion distributions, never decisions).
reference only
- Stanford ODP — automated-facilitation design evidence; hosted limited-alpha, no public source.
- Dembrane Echo — capture/synthesis workflow design; BSL licence and AI-artefact boundary constrain integration.
unresolved
- Panelot — hosted service; no public source/licence. Credible only via Sortition Foundation engagement.
- OpenDLP seed/witness model — undocumented at retrieval; re-check before CA-A1.
- Dembrane Echo API/privacy surface — unverified.
- groupselect vs opendlp lineage — which SF surface is strategic.
The actual missing layer
Not platforms — the governed record spine: eligibility rules, authorised selection specs, witness bundles, statused membership, authorised remits, decision records with dissent, response objects, implementation commitments, exclusion events, AI-artefact labels, nesting relations. Every one is thin engineering but thick institutional meaning. This is the connective layer the CA-R1 landscape found filled by practitioner labour — CA-R3 confirms no assessed product now supplies it.
CA-R2 requirements still poorly served
013–015 (witness/reproducibility packaging), 019 (retention rules), 025–030 (statused membership), 035–037 (authorised-agenda objects), 053 (dissent objects), 061–065 (response/implementation chain), 072 (exclusion events), 080–084 (authority-artefact + AI-boundary), 090–092 (nesting). See REQUIREMENT-COVERAGE.md.
Shared Cicada foundation clarity gained
- Dual-attribution events (actor + authority) and authority artefacts as first-class objects are foundation-level needs, not Assemblies quirks.
- Signed export/reproducibility is a foundation capability.
- Permission≠authority separation needs expressing in the permission system itself.
Assemblies-specific capabilities that remain unavoidable
- Population/frame/spec/run/ceremony chain; cohort/rotation; remit objects; recommendation+dissent; response+implementation chain; exclusion events; nesting.
Questions CA-A1 must resolve
- Composition posture: records-with-channels (D) vs coordinator (A/B) — the framing decision.
- Engine consumption: wrap sortition-algorithms (
adapt) vs consume OpenDLP (integrate) — GPL boundary either way. - Canonical store shape for the record spine (still no schema decisions).
- Which public-input platform, if any, v1 needs.
- Offline member-lifecycle mechanics — unserved everywhere; design or method?
- Personal-data custody boundary between Cicada and operators.
Is a Cicada-native selection engine justified?
No. The published engine satisfies the fairness/optimisation requirements. Cicada’s unavoidable selection-side work is the specification artefact + ceremony/audit packaging + lifecycle joins — thin layers around an existing engine, not an engine.
Smallest plausible Cicada Assemblies technical responsibility
The governed record spine + ceremony layer: eligibility/frame/spec records, witnessed run packaging (consuming an existing engine), statused membership/roster history, authorised remit/evidence/decision records with dissent, the response→implementation chain, and labelled seams to optional external participation tools. Everything else is integration or method.
Completion state
All nine named systems assessed with pinned evidence; licences resolved or marked unresolved; 68 CA-REQs mapped; composition options compared without choosing. CA-R3 TECHNOLOGY / REUSE ASSESSMENT COMPLETE — AWAITING OPERATOR ACCEPTANCE.