What is partition-tolerant renewal, in one sentence?+
A way for a severed agent to keep operating without either halting or self-renewing: collect signed attestations from a Byzantine supermajority of peers, each independently verifying the request extends existing authority without widening it, within a hard ceiling.
Why can't the agent just re-issue its own credential?+
Because that collapses the security model. A node that can renew itself needs no issuer, and therefore answers to none — a single compromised node keeps its own authority alive forever. The requester's own signature never counts toward quorum.
Why not just halt at expiry until the issuer is reachable?+
Because it turns every network partition into a weapon: an adversary who can keep a link down for one credential lifetime disables an agent without touching it — a denial-of-service primitive with a known activation delay.
What is the Renewal Chain?+
The atomic object: an ordered sequence of RenewalProof entries, each a quorum-attested extension of the one before it, independently re-verifiable by any relying party without trusting whoever assembled it. Reference implementation: sm-renewal.
Can the quorum widen an agent's scope while it's renewing?+
No — structurally, not just by policy. No field in the wire format is capable of carrying a wider scope claim; the original scope is copied verbatim across every round, and every attester and relying party checks it stays byte-identical.
Does this fix revocation during a partition?+
No, and it does not claim to. If the issuer revokes a credential while the agent is severed, that revocation cannot take effect until the partition heals — no protocol can deliver a decision to an unreachable node. This mechanism bounds the exposure (the ceiling) and prices it (mandatory revocation resync on reconnection); it does not eliminate it.
What happens if mesh clocks disagree?+
The disagreement tolerance is provisioned mesh-wide and mechanically enforced, not merely assumed: it tightens the ceiling itself, so clock drift can never talk the mesh past the true bound, and any round whose attestations disagree on their own timestamp by more than that tolerance is refused together. What the protocol can't do is manufacture agreement between clocks that have actually drifted apart — keeping them roughly synchronized during a partition is still the mesh's own responsibility.
Is this a WIMSE-specific mechanism?+
No — WIMSE is where the reference implementation happens to plug in first, not the boundary of what this is for. The protocol treats the credential it extends as an opaque, hashed reference: it works identically whether that credential is a WIMSE Workload Identity Token, a SPIFFE/SPIRE SVID, an OAuth-style bearer token, or a physical-AI mission authority issued to a satellite, robot, or vehicle before it went out of contact.
How does this fit the NANDA ecosystem?+
NANDA's discovery layer (the Index, AgentFacts) answers who an agent is and where; identity systems answer what it's authorized to do. Neither has ever specified what a healthy, authorized agent should do when it's cut off mid-task. Partition-tolerant renewal is that missing cross-cutting layer — see What is Partition-Tolerant Renewal for the full picture. A settled Renewal Chain is a natural evidence artifact for registries such as the NANDA Index and its AgentFacts, regardless of which identity system issued the underlying credential.
Is this open source / an open standard?+
The reference implementation (sm-renewal) and the paper and Internet-Draft are complete and tested, in the same family as sm-arp and other Stellarminds.ai primitives — see Research & Drafts.