AirJIT’s protocol claims are not marketing adjectives. They rest on a small set of formal ideas: determinism of execution, graph-structured capacity, and a hard split between optimization (which may be stochastic or heuristic) and state transition (which must be reproducible).
Deterministic execution
Let a logistics state be a tuple of asset identity, custody, location class, and commercial obligation flags. PAEE defines a transition function δ(s, e) = s′ where e is a protocol event with attested inputs. Given the same s and e, every compliant implementation must compute the same s′. There is no random branch inside execution.
The Global Logistics Graph
The GLG is a directed, attributed graph G = (V, E). Vertices encode locations, Hosts, warehouses, and other nodes. Edges encode services and capacity (airline leftover kg is a special case of an edge attribute). ACE searches G for feasible walks. Searching G is not the same as mutating custody on a DLA.
Advisory versus committed
ACE may score walks with costs, latency, and reliability estimates. Those scores can use machine learning. They are recommendations. Commitment happens only when participants accept and PAEE applies δ. This is the mathematical content of “AI does not silently rewrite state.”
Settlement as a separate algebra
Commercial amounts live in J1USD. Escrow release is a function of attested events, not of JIT spot price. Mixing security-token volatility into δ’s money fields would couple unrelated risk processes. Dual-asset architecture is an accounting invariant, not a branding choice.
Oracles
An oracle supplies a typed observation o with provenance. PAEE may require o as a precondition of e. The oracle does not itself compute s′. That keeps external data as a bridge, not a mutator.