Summary
Decide how guardian sets are represented, how they participate in cancel handoff (see Issue #___ cancel/liveness RFC), and how rotation works without becoming a silent takeover vector.
Problem
Guardians are often proposed as the party that can veto a false dormancy escalation. If rotation is instant or single-key, a compromised owner or social-engineered guardian change can capture veto rights or block legitimate handoffs without going through the same delay as the original policy commit.
This issue depends on cancel/liveness semantics (Issue #___) but can be discussed in parallel for data-model choices.
Recommended direction (open for discussion)
- Guardians are explicit in the manifest (addresses + weights or M-of-N), hashed in
policyRoot.
- Cancel handoff requires guardian quorum only when the manifest enables it; owner-only cancel remains for simpler profiles.
- Rotation of guardians:
- Preferred: changes require a delay (timelock aligned with policy commit norms) or a new policy version bump with cooling-off before new guardians can cancel.
- Discouraged: instant replace guardian set without owner delay (abuse path).
- Events for add/remove/rotate so indexers and heirs see governance of the veto set.
Discussion welcome: guardian = multisig contract vs EOAs; institutional guardians; recovery when guardian keys are lost.
What we need to decide (discussion checklist)
What will be affected
- Cancel/liveness RFC (Issue #___)
- Policy manifest (
guardians, quorum fields)
- Vault roles and events
- Trust boundaries and README “who can lie / censor” table
- UX copy for non-legal guardian responsibilities
Target end behavior
- Users understand who can veto and on what timeline.
- Rotating guardians is visible and delayed, not a backdoor.
- Lost guardian keys have a documented recovery path (e.g. owner commit + timelock, or policy version).
Suggested acceptance criteria
Summary
Decide how guardian sets are represented, how they participate in cancel handoff (see Issue #___ cancel/liveness RFC), and how rotation works without becoming a silent takeover vector.
Problem
Guardians are often proposed as the party that can veto a false dormancy escalation. If rotation is instant or single-key, a compromised owner or social-engineered guardian change can capture veto rights or block legitimate handoffs without going through the same delay as the original policy commit.
This issue depends on cancel/liveness semantics (Issue #___) but can be discussed in parallel for data-model choices.
Recommended direction (open for discussion)
policyRoot.Discussion welcome: guardian = multisig contract vs EOAs; institutional guardians; recovery when guardian keys are lost.
What we need to decide (discussion checklist)
proveLivenesson behalf of owner (usually no)What will be affected
guardians, quorum fields)Target end behavior
Suggested acceptance criteria