[Agentproto] Re: AgentProto implementation evidence: Cross-domain dialog recovery after mediator failure (B5-2)

[Nikhil D <[e-mail]>]

Dear Nicholas,

Thank you for taking the time to review the repository and for identifying
this additional acceptance case.

I agree that an Agent A process restart between dispatch and
acknowledgement is an important boundary to test, particularly when the
original downstream execution may still complete.

I will work on a reproducible process-restart scenario using the same
durable operation ledger, with checks for preserved operation identity and
payload binding, no duplicate downstream effect, and explicit unresolved
state until reconciliation establishes the outcome.

I will also retain the downstream atomic-fencing assumption and clearly
distinguish this test from cross-operator authentication or deployed
protocol interoperability.

Once the implementation and tests are complete, I will share the updated
revision and results with the working group.

Thank you again for the constructive feedback.

Best regards,
Nikhil Reddy Donapati


On Mon, Oct 12, 2026 at 12:58 AM Nicholas Templeman <[e-mail]>
wrote:

> Hi Nikhil,
>
> I reviewed the README and the operation-ledger and reconciliation tests at
> revision a940acea0f3e3394e72e4e262e21fe912d40d4d3. I have not run the
> prototype. From that source review, the failover scenario looks useful as
> implementation evidence, particularly the distinction between recovering a
> dialog binding and settling an operation already in flight.
>
> One additional acceptance case would cross the Agent A restart boundary
> that the README identifies as unmodelled: persist an operation as SENT,
> restart Agent A in a new process against the same durable ledger before its
> acknowledgement is recorded, and let the original downstream executor
> finish late while recovery reconciles the outcome. The checks would be
> preservation of the original operation identity and payload binding, at
> most one downstream effect, and an explicit unresolved outcome until
> reconciliation can establish what happened. Resuming the dialog should not
> itself create a fresh operation.
>
> The existing tests already cover lost acknowledgements, late failures and
> fenced late executors. This case would establish whether those invariants
> survive the participant process restart as well. The downstream
> atomic-fencing assumption would still need to remain explicit; passing this
> case would not establish cross-operator authentication or a deployed
> protocol.
>
> Best,
> Nicholas
> _______________________________________________
> Agentproto mailing list -- [e-mail]
> To unsubscribe send an email to [e-mail]
>