---
type: "Architecture"
title: "The federation layer (the missing plane)"
description: "The collaborative-learning plane that lets institutions improve a shared model without centralising data."
tags: [architecture, federated, stack]
generated: { by: human:crpage, at: 2026-07-09T09:44:00Z }
status: stable
---

# The missing layer
Most AI stacks combine three layers — data platforms, application workflows, and model inference — but lack a **federation layer** that coordinates model training, evaluation and analytics *across organisational boundaries*. [MCP](../standards/mcp.md) and [Arazzo](../standards/arazzo.md) help with interoperability at inference and workflow time; [federated learning](../federated/federated-learning.md) frameworks solve interoperability at **optimisation and governance** time.

# Why it matters for public services
Councils, NHS bodies and ALBs often need more than a tool-calling assistant: a controlled way for multiple data holders to contribute to shared intelligence without merging raw data — implying client selection, aggregation, privacy controls, node identity, auditability and federation governance.

Relates to: [Flower](../frameworks/flower.md), [Cross-device vs cross-silo](../federated/cross-device-vs-cross-silo.md), [The missing collaborative-learning layer](../document/themes/missing-collaborative-learning-layer.md).

# Terms
Glossary terms used here: [Audit trail](../glossary/audit-trail.md).