SPECTRE · research design
Local delivery evidence can inform a next-hop choice. The EBM hypothesis must beat simpler baselines and survive failure, stale signals and misleading peers.
SPECTRE overview SPECTRE is the proposed shared node and transport network. MESH is the many-to-many routing pattern that runs across it. The routing hypothesis on this page is that local route evidence can help a node choose a next hop; it does not replace the ledger’s agreement protocol. Each participating node would discover available neighbours, score candidate next hops for a packet class and destination, forward the packet, then record the local outcome. Different traffic may value different things: a voice packet may prefer low delay; a large model download may prefer throughput and cost. Nodes need not agree on one global “best” path. Each node can make a bounded local choice from its own observations, then the next node continues the route. Alternate paths and a conservative fallback are part of the design problem. Research target: the SPECTRE repository holds a routing skeleton, not a deployed network service. The diagram does not represent measured performance. The research direction is a local energy-based model (EBM) that scores candidate neighbours or paths for a traffic class from observations the node can actually make. A source-level EBM and router prototype exist in the repository; they do not establish a working multi-node network, per-node learning at scale or measured routing gains. Reynolds’s flocking model shows how simple local-neighbour rules can produce coordinated motion. This is an analogy for what a node can observe, not a routing algorithm or proof. The EBM objective, update rule, reward, decay, convergence and stability still need specification and comparison with simpler baselines. Packet content should not be a routing feature. Even with encrypted payloads, a forwarding node may observe source or destination identifiers, timing, size and path choices; the threat model must say what is exposed and for how long. Peer-reported performance is an untrusted observation, not proof. Compare MESH application-routing policies on the same workload before adding a learned route model. Random-peer forwarding is one application-traffic baseline; Hashgraph consensus gossip is a separate protocol with its own assumptions. Report delivery ratio, p50/p95/p99 latency, throughput, route changes, control overhead, cost and energy per successful packet, and recovery time after node churn. Simulation is a first comparison, not network evidence. Reproducible experiments need a described topology, workload, node failures, adversarial behaviour, baseline configuration and independently checkable results. Skyzai’s consensus network, messenger, VPN, CDN and inference cloud are not presented as live services here. These chapters explain a research design. An animation, source specification or local website test does not establish a running network.SPECTRE · the shared node network
What a node would do
Why local routes
A small model learns from route outcomes
Learn from neighbours, not a global map.What the research has to prove
Continue through SPECTRE.
Context: Routing & learning.