Agentic Activity & Evidence Scenarios

Autonomous Payment

Initial observation: payment.release for USD 186,400. Local policy requires verified authority above USD 50,000. Credential VERIFIED; authority UNRESOLVED. Policy decision: ESCALATE. Added authority evidence verifies a USD 250,000 limit. Updated state permits policy re-evaluation while preserving the first decision. Execution remains NOT_STARTED; no settlement is established.

Request: USD 186,400 → Credential: VERIFIED
  → Authority: UNRESOLVED → Local policy: ESCALATE
  → Authority evidence: USD 250,000 VERIFIED
  → Policy re-evaluation available
Execution: NOT_STARTED | Settlement: UNAVAILABLE

Delegated Authority Failure

An authenticated actor presents no sufficient delegation evidence. Identity may be VERIFIED while authority remains UNRESOLVED and policy denies execution.

Identity: VERIFIED → Authority: UNRESOLVED
  → Local policy: DENY → Execution: NOT_STARTED

Execution Mismatch

The requested action and provider-reported execution differ. APP preserves the discrepancy and marks the outcome claim CONFLICTED.

Requested scope A → Provider reports scope B
  → Comparison evidence → Outcome claim: CONFLICTED

Cross-System Correlation

A runtime activity and provider execution are CORROBORATED by independent identifiers and timing without being represented as deterministically linked.

Runtime observation + Provider observation
  → Independent corroboration → CORROBORATED
Not automatically DETERMINISTICALLY_LINKED

Conflicting Evidence

Two trusted sources disagree about principal or authority. APP marks the claim CONFLICTED and retains both sources for review.

Source A: authority X + Source B: authority Y
  → Claim: CONFLICTED → Both records preserved

Authority Revocation

Previously verified authority becomes REVOKED. Current policy state changes while the earlier valid interval remains in the evidence trail.

Authority: VERIFIED → Withdrawal observation
  → Current authority: REVOKED
Earlier observations remain preserved

Tool Invocation

A runtime asserts intent, a gateway observes the request, and the tool records execution. APP preserves three observations rather than treating the request as proof of execution.

Runtime: intent ASSERTED → Gateway: request OBSERVED
  → Tool: execution evidence → Independent review

Unknown External Agent

At an Outside-In boundary the actor may remain UNRESOLVED while credential, action and local policy are available.

External request → Protected boundary
Actor: UNRESOLVED | Credential: VERIFIED
Action request: OBSERVED

Inside-Out + Outside-In

Runtime intent and gateway observation strengthen the same activity representation while preserving separate trust domains.

Controlled runtime → Source observation A
Protected gateway → Source observation B
  → Evidence-backed correlation

Human Escalation

Policy requests human review when evidence is insufficient. The review outcome becomes a new observation, not a rewrite of the original state.

Evidence gap → Local policy: ESCALATE
  → Human review → New observation
Original assessment preserved

Policy-Regime Difference

The same evidence can lead to different local decisions in different policy domains without changing the underlying evidence state.

Same evidence → Policy domain A: ESCALATE
Same evidence → Policy domain B: DENY
Evidence state remains independent of enforcement