Peers_Subsystem

Peers Subsystem

brain/peers/ implements the outside observers described in Peers: entities that read Orrin's state from the outside and, when their wake conditions fire, push signals into the next cycle rather than mutating state directly. They flow through the same signal_router (brain/think/signal_router.py) as every other signal, so they nudge attention without ever issuing commands.

The five peers

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

Mechanics

• peer_base.py — the base class: a wake condition (cheap, evaluated against current state) plus a proposal generator that emits signals when awake.
• peer_registry.py — registration and scheduling; peers register themselves in the world model / relationships on first wake, so Orrin's self-model knows they exist.
• Peers are deliberately advisory: their output is salience, not control. A peer can make the workspace competition care about something; it cannot force an action.

Writing a custom peer

Inherit from the base class in peer_base.py, implement the wake condition and proposal generation, and register it in peer_registry.py. Keep wake conditions cheap (they run often) and emit signals, not state mutations. Walkthrough: Adding a Custom Peer.

Code pointers

• brain/peers/ — all of the above
• brain/think/signal_router.py — where peer signals join everything else