Spaxiom: Global Experience Fabric for Agents - Part 9 of the Technical Series

About This Section

This section describes how Spaxiom scales from local site deployments to a global experience fabric that aggregates events across many sites. Unlike raw sensor data, Spaxiom's typed event schema enables cross-domain learning and multi-site coordination.

You'll learn about the shared event ontology across hospitals, warehouses, retail, and offices; how events aggregate into a heterogeneous graph structure (sites → event types → model training pipelines); and the three primary uses of the fabric: supervised learning, unsupervised pattern mining, and world-model pretraining for agents.

4. A Global Experience Fabric for Agents

From local deployments to an "experience fabric"

A key goal for Spaxiom is not just to orchestrate sensors within a single site, but to define a shared ontology of events across many deployments:

Hospitals: gait stability, falls, equipment movements, staff–patient interactions.
Warehouses: forklift trajectories, pallet flows, near-misses, congestion.
Retail: dwell times, queue patterns, shelf interactions.
Offices / campuses: occupancy patterns, informal collaboration zones, comfort signals.

Each deployment runs a local Spaxiom + INTENT stack that yields typed events:

// Example: cross-domain event schema
{
  "type": "CrowdFormation",
  "site_id": "retail-347",
  "zone": "checkout-lane-3",
  "start_time": "2025-11-05T13:20:00Z",
  "end_time":   "2025-11-05T13:27:00Z",
  "peak_occupancy_pct": 45.3,
  "avg_wait_time_s": 190.0
}

{
  "type": "GaitInstability",
  "site_id": "hospital-5f",
  "zone": "ward-b-hall-2",
  "timestamp": "2025-11-05T09:13:22Z",
  "stability_score": 0.23,
  "recent_steps": 28,
  "assistive_device": "walker"
}

With a consistent schema, these events can be aggregated into a global experience fabric.

Graph view

We can model this fabric as a heterogeneous graph G = (V, E):

Vertices V include:
- Sites v_site
- Zones v_zone
- Event types v_etype
- Event instances v_event
Edges E capture relationships:
- v_site → v_event: event occurred at site.
- v_event → v_etype: instance-of relationship.
- v_zone → v_event: event occurred in zone.
- temporal edges between events (before/after/overlaps).

Figure 2 (Experience Fabric Graph)

Sites

warehouse-A

hospital-B

store-C

→ → →

Shared Event Types

CrowdFormation

QueueOverflow

FallEvent

→ → →

Model Training

World Models

Policy Networks

Safety Monitors

A tri-partite graph showing: Left: many sites (warehouse-A, hospital-B, store-C); Middle: shared event types (CrowdFormation, QueueOverflow, FallEvent); Right: model-training pipelines consuming these event streams.

This fabric can be used in at least three ways:

Supervised learning on top of events (e.g., predict FallEvent from preceding patterns).
Unsupervised pattern mining (e.g., cluster recurring failure modes).
World-model pretraining for agents: training on event sequences as "experience tokens" rather than raw sensor tokens.