// Section 2.2 · Concepts

The coordination layer in detail

1 min2.2Concepts

// 2 of 4 · network anatomy

Coordination is centralised at launch, by design.

// 2.2 · the coordinator in detail · four logical services

The architecture commits to behaviour, not topology.

Coordination is centralised at launch. That is a deliberate trade-off: ship a working network with one moving part, rather than a partial multi-validator system on day one. The coordinator is described honestly as a single point of failure until decentralised validator quorums land.

Four logical services

// intake · segmenter · scheduler · validator

// intake apiIntake API

Accepts requests, verifies $PRLX payment, enqueues for routing.

// segmenterSegmenter

For genuinely parallelizable workloads, splits the task graph into independent sub-tasks. AI inference is dispatched whole, not segmented.

// schedulerScheduler

Matches each unit of work to a node using the four scheduling signals (§IV).

// validatorValidator

Reconstructs Proof-of-Execution commitments, re-dispatches a sampled fraction, settles payment.

Topology is deployment-time

// behaviour is canonical · topology is a deployment decision

// OPTION ASingle process

All 4 services in one binary, one host. Suitable for a single-host deployment.

// OPTION BSingle cluster

Each service its own pod, shared cluster. Launch production scale.

// OPTION CMulti-cluster / multi-region

Services federated across regions. Planned, when quorum validators come online.

Topology progression

// the coordinator becomes a quorum

AT LAUNCHSingle coordinator

Documented single point of failure. Sampled re-dispatch is the only honesty check.

PLANNEDMulti-instance + validator quorum

Coordinator runs as a multi-instance service. Decentralised validator quorum overlays the re-dispatch layer.

LATERFederated quorum + L1+L2 settlement

Coordinator quorum federates across regions. On-chain dispute resolution for validator disagreement.

Caveat·Honest trade-off at launch

The coordinator is a documented single point of failure at launch. The design goal is to ship a working network with one moving part rather than a partial multi-validator system. Decentralised validator quorums are a planned later addition.

// Reading path

Network Anatomy · parent hub Threat model · single-point-of-failure context §16.2 Decentralisation progression