Summary

  • Girish Konda's 19 October individual Internet-Draft calls a fetched input stale only when its age is greater than the specified freshness bound. Its TC-2 test says equality remains inside the limit.
  • Wes Jackson's updated 28 October individual draft bars use of the same input at or past its horizon. At the boundary, a verifier must fetch again; if a required source does not answer, it records a transient refusal.
  • Konda's model classifies evidence, leaving the final decision to local policy. The conflict concerns the state of the cached input, not a Konda rule requiring admission.

One instant, two ledgers

Suppose an agent's issuer standing was confirmed at time A. The relying party permits that source result to live for M more units of time. At A+M, the issuer-standing source cannot be reached. A verifier still holds the last affirmative answer. May it count that answer as established evidence for this decision?

Konda's Preserving Evaluation State in Agent Protocol Decisions, revision 02, says the age is stale when it exceeds the bound. Its definition expressly includes equality. Test case TC-2 makes the clash visible: an input whose age equals the bound stays within it, while Jackson's verifier draft already treats it as past the horizon. Konda's wider point is that a negative answer, a missing answer and an old answer are different facts. Flattening them into “denied” sends callers and incident teams in the wrong direction.

Jackson's Verifier-Side Evaluation Semantics for Delegated Authority Chains, now at revision 05, draws the line one instant earlier. Section 3 says a verifier must not use a fetched status, key set or record set at or past its as-of time plus the pinned maximum age. It must obtain the input again. Appendix A.3 supplies the operational test: at or after A+M, a new affirmative standing result can support admission; a source that does not answer produces a transient refusal, recorded as an evaluation that could not finish. It is not a finding that the issuer lost standing.

Those words govern more than a stopwatch. Under Konda's state model, the cached value at equality remains established, although local policy may still refuse the call. Under Jackson's rule, that value cannot establish the required input; an unreachable source means the verifier refuses for now. One operator may diagnose a policy decision on known standing. Another may diagnose a dependency outage. A retry that is useful in the second system may be pointless in the first. The difference appears precisely where ordinary tests often use only “before” and “after.”

Both texts are active individual Internet-Drafts. Their Datatracker pages show no adopted IETF standard or RFC for either proposal. The concrete editorial stake is therefore a rule-writing choice: if these semantics are meant to interoperate, the equality case needs a shared test and a shared name for the outcome. Konda already names Jackson's earlier draft in TC-2. Jackson's newer -05 still uses the stricter comparison. A later revision or joint conformance case that resolves equality and source unavailability would change this assessment.

Sources