A.2.5:12 - SoTA-Echoing
| Current or mature line | What it contributes | Practical use in A.2.5 |
|---|---|---|
| W3C SCXML 1.0, a mature 2015 Recommendation rather than current competitive SoTA | Explicit states, parallel regions, guarded transitions, events, and executable state-machine semantics. | Keep statecharts available when the subject-domain model needs them, but type them as mathematical or description lenses rather than the world-side relation occurrence or universal Method order. |
| Esparza and Fischer, Runtime Verification for LTL in Stochastic Systems, 2025 | Runtime monitoring distinguishes true, false, and inconclusive results; finite observations do not settle every temporal property. | Treat incomplete evidence as unresolved for the relying use, preserve the predicate’s temporal reading, and do not close an occurrence merely because a finite evidence path is silent. |
| Cedar Policy Language current reference | Fine-grained decisions evaluate a concrete principal, action, resource, current attributes, and request-time conditions rather than a system-role label alone. | Require the system performing consumer decision Work to combine current assignment, exact predicate, state window, and action-specific relations. Keep this as an implementable software specialization rather than the ontology of every assignment state. |
| Zuvic, Capability Gates Are Not Authorization, 2026 preprint | A current agent-framework audit distinguishes exposed capability from per-call, value-sensitive authorization and reports fail-closed enforcement experiments. | Keep capability in A.2.2 and require the consumer to evaluate the concrete state and action claim before side effects; do not infer authorization from tool exposure. The empirical scope remains the audited software frameworks. |
| Liu et al., A Framework for Formalizing LLM Agent Security, 2026 preprint | Task alignment, action alignment, source authorization, and data isolation require runtime checks over the current task and action. | In agentic cases, require the consumer’s governing claim to name the current task and action relations; A.2.5 supplies only the exact state relation and assertion form, while A.10 supplies only the evidence-use relation; the applicable evaluation or assurance pattern separately establishes any reliance posture. |
A.6.REL, A.2.1, A.19, A.2.4, and A.10 | FPF already separates relation obtaining, occurrence identity, assignment episodes, characteristic-space predicates, assertions, and evidence use. | Use §4.3 to distinguish state episodes; record an assertion and evidence use only when the receiving use requires them. |
The sources’ transferable contribution is bounded: current action decisions need exact participants and predicates; temporal monitoring can remain unresolved; capability and action admission differ; and state-machine notation is optional modeling machinery.