Theory to practice — CA-R1 synthesis
Updated: 2026-10-05. Central synthesis linking the theory/methods records to the case corpus. Research findings only — no architecture, no policy decisions.
Claims strongly reflected in practice
- Deliberated judgement is real and condition-dependent: corpus assemblies that paid for time, evidence and facilitation produced considered collective outputs (CA UK, Scotland, Ireland, SRC-0038/0044/0017; empirical review SRC-0106).
- Response obligations determine impact: the single strongest practice–theory alignment. Statutory response (Scotland, SRC-0043) and justified-rejection (Ostbelgien, SRC-0021) produce institutional uptake; courtesy responses decay (Leeds, SRC-0047). This matches the systems claim that legitimacy is a chain, not a property of the room (SRC-0111).
- Stratified selection corrects invitation bias on measured variables: CA UK’s strata + overweighting produced population-close composition (SRC-0038) — stratification works as theory predicts, within its limits (SRC-0109).
- Support provisions are representativeness infrastructure: where payment/care/access were funded, harder-to-reach strata were seated (SRC-0038, SRC-0101).
Claimed principles inconsistently implemented
- Equal probability of selection: honoured at invitation (weighted exceptions), not at sitting — acceptance filters reintroduce population gradients everywhere opt-in exists (SRC-0120, SRC-0038).
- Transparency: selection algorithms published (CA UK code, OpenDLP docs) but audit ceremonies (seed, witness, logs, replacement protocol) are under-documented across the corpus (gap, SRC-0079, SRC-0123).
- Dissent capture: Leeds tiers and Irish minority reports exist, but most cases publish only outputs — strength-of-support recording is inconsistent (SRC-0047, SRC-0017).
- Deliberative quality measurement: no corpus case publishes a deliberative-quality metric; evaluation is experiential (SRC-0119).
Where practice conflicts with theory
- Attitudinal stratification: theory treats opinion as emergent deliberative input; CA UK made it a selection criterion — defensible (mirroring opinion landscape) but risks engineering outcomes; Brussels’ rules + GDPR-prohibition approach is the stricter alternative (SRC-0038 vs SRC-0031/0032). Unresolved.
- Protected vs public deliberation: transparency theory wants visibility; domination/performance theory wants protection — Brussels legislates protection (SRC-0030). The corpus resolves by publishing method and outputs, protecting process.
- Consensus as ideal vs as practice: practice hybridises — consensus-seeking as norm + ballot closure + published support levels (SRC-0044, SRC-0047). Pure consensus is neither required nor evidenced as superior.
Design choices that are inherently normative, not technical
- Strata selection and weighting (who counts as relevant difference) — SRC-0109.
- Attitudinal/political criteria — legitimacy contest.
- Invitation weighting (equal raw probability vs corrected representativeness).
- Agenda authority (who may place the question).
- Response/implementation obligations (what the institution owes).
- Transparency vs protection of the deliberative space.
- Delegation vs fresh sortition in multi-level designs.
Each of these can be recorded but must not be decided by software defaults — they belong to institutional governance.
Matters that cannot legitimately be delegated to software
- Choosing strata and their political rationale.
- Choosing expert/advisory composition and evidence balance.
- Facilitation judgements in the room (influence is human and must remain auditable to humans).
- Agenda-setting authority.
- The response obligation itself (an institutional commitment, not a mechanism).
Software can implement audited draws, publish composition, track responses — but the legitimacy-relevant decisions are institutional (SRC-0111; cf.
CAPABILITY-LANDSCAPE.mdconnective-layer finding).
Required flags — critical issues
- Equal chance vs stratified representation: tension recorded; leximin manages within-pool fairness, cannot repair frame/response bias (SRC-0078, SRC-0109).
- Self-selection after random invitation: universal, partially corrected by strata/overweighting; never eliminated (SRC-0120).
- Attitudinal stratification: CA UK vs Brussels = contested governance; unresolved.
- Participant barriers: support = representativeness infrastructure; under-documented outside CA UK.
- Facilitator power: real, managed not eliminated; least auditable stage.
- Expert/evidence gatekeeping: CA UK’s three-tier governance is the benchmark; conflicts-of-interest disclosure uneven.
- Agenda-setting power: least-studied consequential power; five corpus models (
agenda-setting.md). - Commissioner influence: maximal at remit and expert-selection stages (SRC-0100).
- Response vs implementation: only response is ever obligated; implementation tracking is the corpus’s thinnest stage (Paris droit de suite is the best instrument, SRC-0027).
- Accountability of unelected bodies: runs through method transparency + bounded mandate + response duty, not elections; permanent councils add rotation+memory (SRC-0118).
- Legitimacy of permanent sortition institutions: theoretically supportable via rotation/bounded mandate; can become unaccountable incumbency — controls are design, not automatic (SRC-0109).
- Relationship to elected institutions: hybrid coupling (Brussels 12 MPs + 45 citizens) is the corpus’s working answer; pure sortition chambers are untested (SRC-0030, SRC-0125).
- Privacy vs transparency: special-category strata prohibited (Brussels) vs consent-based (UK); small-cell publication risk unresolved (SRC-0032, SRC-0122).
- Local/regional/national nesting: theory ahead of practice; equal-probability (fresh draws) vs linkage (delegation) is the core trade-off (
nested-and-multi-level-assemblies.md).
Concepts Cicada must eventually be able to represent explicitly
(Preliminary research synthesis — not a data model.)
- Sampling frame type + coverage gap.
- Invitation funnel (sent/registered/selected/seated) + weighting.
- Strata table + who chose it + under what legal constraint + publication status.
- Selection algorithm + seed/witness/audit artefacts + replacement protocol.
- Support provisions per barrier type.
- Agenda authority model.
- Evidence-governance structure + disclosure.
- Facilitation method + oversight.
- Deliberation-vs-decision method separation + dissent recording.
- Mandate type + response obligation + tracking instrument.
- Permanence/rotation/institutional memory.
- Nesting relationships (if any) + delegation vs fresh-sortition mode.
- Mass-input channel + its bias profile (distinct from membership).