Re: Four open questions on the Agentic Web report (2026-09-24): preimage defense, address assurance, trust leases, statement granularity

Hi Lars, Xinghua, Chris, Lee, Tymofii, and colleagues,

Lars, understood. We should post the proposed field list and both finish-or-stop contracts here for your written review, then record any changes agreed during the session.

Xinghua, I support giving resume_basis an explicit decision point. We should distinguish its purpose from its representation, and identify the evidence a receiver must verify before resuming. The field’s presence or a declared delegation state alone should not establish authority.

Chris, thank you for offering the harness. Once the contract and implementation versions are pinned, we can identify which cases it exercises and compare the recorded outcomes.

For the first test family, I suggest we keep the agenda focused on stopping semantics, independent resource observations, and PAUSE receipts, with new-call admission checked separately. We can then circulate the proposed expected outcomes for review before execution.


Best regards,

Larry Lewis

Founder, SovrLabs

Chair, W3C Agent Trust Protocol Community Group




From: AlgoVoi <chopmob@gmail.com>
To: <public-agentprotocol@w3.org>
Cc: <llewis@sovrlabs.com>, <lee@aicent.com>, <abisheakp197@gmail.com>, <tima@aeoess.com>, <zhaoxinghua07@126.com>
Date: Fri, 30 Oct 2026 08:50:32 -0500
Subject: Re: Four open questions on the Agentic Web report (2026-09-24): preimage defense, address assurance, trust leases, statement granularity



Colleagues, 
 
Xinghua's proposal for a resume basis is the right shape, and I have implemented it against a conformance harness so the claim can be tested rather than argued. 
 
The first point is the one that convinced me. Verifying a signature tells you who signed. It cannot tell you whether the condition that caused the halt still holds, so a run reopened on evidence and a run reopened on exemption end in records a reader cannot separate. Naming the basis on the receipt fixes that without adding a state or removing one. 
 
Both boundaries are implemented as proposed. An exemption cannot cover a revoked authority, because revocation is the issuing authority's decision and a manager on the executing side cannot undo someone else's revocation. An exemption cannot cover an undetermined authority, because that is an absence of information, and a signature cannot convert we do not know into we know. 
 
I also agree with the second point, that irreversibility belongs in the missing transition rather than in a terminal state. Our own code made the case better than the argument did. The resume transition was not declared in the state machine at all, so it could not be withdrawn by anything. It is declared now, and where a revocation is final under the governing regime the chain simply never reaches the resumed state, which leaves the state space finite and enumerable. 
 
The harness is below. It is Apache 2.0, it has no runtime dependencies, and it checks an implementation against the specification rather than reimplementing it. 
 
https://github.com/chopmob-cloud/algovoi-ube  
 
Two limits, stated rather than glossed. It does not implement Byzantine consensus, which the specification requires for high value multi agent actions, so that gap is recorded instead of bridged. And the five values I have used on the wire are provisional until they are reconciled with Section 8 of the draft. 
 
If it would help, I am willing to run it against the reference implementation and report what it measures, rather than what the document says it should. 
 
Best Regards, 
AlgoVoi