// Section 4.1 · Protocol
Capability Match
The binary filter. Yes or no, before anything else.
// 4.1 · capability match · signal 1 of 4
Every routing decision starts with the same question: can this node run this class of work? The answer is binary. Yes means the node enters the candidate pool for the next signal. No means the node is filtered out. There is no partial match, no preference weighting at this stage. The unit of work is the dispatched item: a whole inference request on the live AI path, served end to end by one node, or a sub-task on the general parallelizable-workload path.
Five declared workload classes
// declared at registration · class-based routing is planned
How the filter applies
// work class meets node capability set
req_a3f1 class: ai.inference tier: 3 size: medium
// 2 of 4 nodes survive the binary filter; the next 3 signals run on this pool only
Routing posture
// declarations exist now · class-based routing is planned
Capability declarations are recorded at registration. The scheduler treats every capable node as eligible for every request. Capability is a metadata field, not a routing filter.
Capability declarations remain unchanged. Reputation and uptime get observed per class ahead of the routing transition.
Workload-class optimisation ships. Work class becomes a binary filter at dispatch. Mis-declared capabilities are flagged at re-dispatch reconciliation.