SARB charter
Fifty sortitioned citizens serve cross-polis. Ten seats are technical review seats and forty seats are lay citizen seats. Rotation is six months, replacing twenty-five seats every three months.
SARB
The algorithm answers to citizens.
The Sortitioned Algorithm Review Board approves every matching-weight change, audits fairness, and can halt unsafe algorithm work by supermajority.
Authority
SARB approves algorithm-weight changes, fairness audits, matching-weight publication, and halt decisions. It cannot create ad revenue, pay-to-play matching, or hidden engagement objectives.
Invariant I18
Every matching-weight write must carry a SARB approval id, reason, before vector, and after vector in the matching-weight history.
Current panel composition
Aggregate-only public panel record
This public record shows composition bands only. It does not publish names, addresses, contact details, citizen ids, private votes, or Constellation fields.
| Dimension | Aggregate composition | Rule |
|---|---|---|
| Total seats | 50 citizens | Cross-polis SARB panel |
| Review role mix | 10 technical seats, 40 lay seats | 20% technical, 80% lay |
| Rotation | 25 seats replaced each quarter | Six-month term cadence |
| Privacy | 0 per-citizen public rows | Aggregate publication only |
Invariant I18 ledger
Last 10 algorithm-weight changes
Each row gives the public reason, before vector, after vector, and SARB approval-id. The source ledger is weight-changes.json.
| Date | Reason | Before | After | Approval id |
|---|---|---|---|---|
| 2026-05-26 | Quarterly bridge audit raised anti-bubble protection. | epsilon 0.09, zeta 0.06 | epsilon 0.10, zeta 0.05 | SARB-2026-010 |
| 2026-05-12 | Restored civic benefit after Assembly readback. | delta 0.09, beta 0.26 | delta 0.10, beta 0.25 | SARB-2026-009 |
| 2026-04-28 | Reduced weekly strain after attendance pacing review. | zeta 0.04, alpha 0.31 | zeta 0.05, alpha 0.30 | SARB-2026-008 |
| 2026-04-14 | Balanced compression progress against 24-hour joy. | beta 0.24, gamma 0.21 | beta 0.25, gamma 0.20 | SARB-2026-007 |
| 2026-03-31 | Cross-cluster review returned homogeneity penalty to baseline. | epsilon 0.11, delta 0.09 | epsilon 0.10, delta 0.10 | SARB-2026-006 |
| 2026-03-17 | Newcomer outcomes favored stronger tie prediction. | alpha 0.29, gamma 0.21 | alpha 0.30, gamma 0.20 | SARB-2026-005 |
| 2026-03-03 | Agora review lowered short-term affect dependence. | gamma 0.22, beta 0.23 | gamma 0.20, beta 0.25 | SARB-2026-004 |
| 2026-02-18 | Venue accessibility audit preserved hard constraints and adjusted burnout guard. | zeta 0.06, epsilon 0.09 | zeta 0.05, epsilon 0.10 | SARB-2026-003 |
| 2026-02-04 | First-quarter calibration returned default civic value. | delta 0.11, alpha 0.29 | delta 0.10, alpha 0.30 | SARB-2026-002 |
| 2026-01-21 | Published phase-zero default utility weights. | unpublished phase-zero vector | alpha 0.30, beta 0.25, gamma 0.20, delta 0.10, epsilon 0.10, zeta 0.05 | SARB-2026-001 |
Halt threshold
Two-thirds halt votes finalize a workflow as halted. For a 50-seat SARB, that means at least 34 halt votes.
Halt history
Public halt events: 0 active halts, 0 unresolved halted workflows, and 0 private vote rows published.
Readback boundary
SARB records are aggregate governance evidence. They are not calibration evidence, private citizen data, or BUILD_DONE completion by themselves.