The Averaging Problem

When a centralised controller sees only the mean, local crises become invisible — and the response makes things worse.

Centralized Controller View

100.0

System Mean Stability

Node-level data is hidden — you see only the average.

Cumulative Stability Deficit

Total deficit: 0

0
0
1
0
2
0
3
0
4
0
5
0
6
0
7
0
8
0
9
0
How this works

A shock hits nodes 2 and 7 at time step 20. In centralized mode, the controller sees only the national mean (with noise) and applies a uniform policy with long latency. The mean barely moves, so the response is weak — while healthy nodes are disrupted.

In distributed mode, each node observes its own local state with low latency and applies its own correction. The crisis is contained, and healthy nodes remain stable.

Press Run to see the simulation. In centralized mode, after it finishes, click Reveal Node Details to see what the dashboard hid.

GitHub Discord E-post RSS Feed

Built with open source and respect for your privacy. No trackers. This is my personal hub for organizing work I hope will outlive me. All frameworks and writings are offered to the commons under open licenses.

© 2026 Björn Kenneth Holmström. Content licensed under CC BY-SA 4.0, code under MIT.