Software — Panelot
Classification: stratified-selection engine (hosted academic/practitioner tool; fair panel selection by lot). Updated: 2026-10-05. Evidence labels per RESEARCH-METHOD.md. Fit: unresolved — candidate for later CA-R3 assessment.
Identity
- Project: Panelot — “panel selection by lot”, a not-for-profit system at https://panelot.org (SRC-0078, SRC-0079).
- Origin: academic research system by Flanigan, Gölz, Gupta, Hennig and Procaccia; implementation described by Procaccia as “our open-source system, Panelot” used for 40+ real assemblies (SRC-0077 paper context; SRC-0097-related talk). Brett Hennig is also a Sortition Foundation co-founder — the algorithm lineage is shared.
- Website/service only: a public source repository matching
panelotwas not located in this pass (Open question; see limitations).
Licence and distribution
- Hosted web service; terms not prominently published in retrieved pages.
- Licence: described as open-source in maintainer publications but no licence file was verified (Open question — do not assume reuse rights).
Core purpose
- Observed: select a citizens’ panel that is representative of the population while giving volunteers as equal a selection probability as mathematically possible (SRC-0079).
Selection / sortition model
- Observed (distinct from simple quota sampling): two-step method (SRC-0078, SRC-0079):
- Compute a civic lottery — generate a leximin-optimal probability distribution over quota-compliant panels (maximise the lowest individual selection probability, then the second-lowest, etc.).
- Draw a panel — select a single panel from that distribution (the paper describes a “live uniform lottery” over the computed panels, SRC-0077).
- Inputs: categories CSV (
category, name, min, max) and pool CSV (anonymisedpool_member_identifier+ group membership columns); panel size (SRC-0080). - Fairness objective: leximin optimality over individual selection probabilities — a published mathematical guarantee, not ad-hoc quota fulfilment (SRC-0078, SRC-0079).
- Observed: client-side anonymisation — files are stripped of identifying information in the browser before submission; the server stores only request metadata (time, pool size, panel size), not the pool (SRC-0080). A deliberate data-minimisation design.
Audit / reproducibility
- Observed: the computed lottery (panel list + probabilities) is itself the audit artefact; the final draw can be run as a public live lottery (SRC-0077, SRC-0080). Analysis: this two-step split — publishable distribution → simple witnessed draw — is a strong reproducibility pattern for civic selection.
- The same algorithm family is independently implemented in
sortitionfoundation/sortition-algorithms(maximin, leximin, nash, diversimax, legacy) (SRC-0077).
Identity and privacy
- Observed: requires de-identified pool data; warns operators to check identifiers contain no personal data; minimal server retention (SRC-0080).
API / export / interoperability
- Observed: CSV in/out; no public API documentation found (Open question).
Deployment evidence
- Observed/Claimed: the Nature paper reports the algorithm implementation used to select more than 40 citizens’ assemblies worldwide (SRC-0077). Panelot is the reference web deployment.
Related tools
- newDemocracy Foundation’s browser-based Stratified Random Selection Tool (selection.newdemocracy.com.au): JavaScript implementation offering “simple” and leximin-style algorithms, explicitly modelled on Sortition Foundation’s earlier Python tool; source published (SRC-0081).
Limitations / open questions
- Public source repository and licence for the hosted Panelot service not verified (Open question).
- No participant-management features (no invitations, reserves, onboarding) — selection engine only (Observed).
- Whether the hosted service’s build corresponds to any public code is unverified.
Sources
- SRC-0077 (Nature paper — algorithm + deployment claim), SRC-0078 (service two-step), SRC-0079 (leximin method), SRC-0080 (inputs/anonymisation), SRC-0081 (newDemocracy sibling tool), SRC-0077 also covers sortition-algorithms.
CA-R3 pinned assessment
- Pinned source/ref: current hosted service (SRC-0152); no public source repository or licence could be established.
- Licence: unresolved.
- Maintenance: service is live.
- Deployment model: hosted service; independent self-hosting not evidenced.
- Relevant CA-REQ coverage: strong conceptual fit for CA-REQ-010–014 and quota/fairness parts of CA-REQ-011; no member lifecycle.
- API/integration seam: anonymised CSV input/output and public-lottery workflow.
- Audit/provenance: the two-stage computed lottery then draw is a strong audit-ceremony reference pattern.
- Identity/privacy: deliberate de-identification is a strong reference model.
- Operational burden: low as hosted service; unknown for self-host.
- Material gaps/adoption risks: source, licence, deployment and API remain unverified.
- Provisional classification: unresolved.