Peers

Peers (Outside Observers)

Peers are entities that watch Orrin's state from the outside and, when their wake conditions fire, push signals into the next cycle. They are advisory by construction: a peer can make the workspace competition care about something, but it cannot issue a command or mutate state directly.

The five peers

Peer	Watches for
Architect	Self-modifications — reviews proposed code before it lands
Signal Historian	Chronic control-signal patterns (stuck pressure)
Goal Auditor	Low-quality or unmeasurable goals
Observer	Unproductive loops and repetitive behavior
Reward Auditor	A bandit reward signal that has collapsed to noise

How they integrate

Peers register themselves in the world model / relationships on first wake (so Orrin's self-model knows they exist) and flow through the same signal_router as every other signal. They nudge attention without ever commanding action — the same "propose, don't race" discipline as the rest of the architecture.

For the implementation and how to write your own, see the Peers Subsystem deep dive and Adding a Custom Peer.

Code pointers

• brain/peers/ — the peers and registry
• brain/think/signal_router.py — where peer signals join everything else