The eight-protocol governance stack

The governance stack is eight protocols forming a single escalation chain. It was the first major milestone of the project and it was built before any intelligence sat on top of it, because a system that cannot govern itself has no business being clever.

Sentinel     →  detects anomalies
Serpentine   →  classifies stress state
Aegis        →  contains failures
Atlas        →  rebalances load
Punisher     →  repairs with AI
Phoenix      →  orchestrates recovery
Hyperion     →  autonomous emergency shutdown
Terminus     →  permanent human-authorised termination

Every protocol has a defined role. Every handoff is deliberate. Nothing overlaps — and the non-overlap is a design rule rather than an outcome, because overlapping responsibilities are where escalation chains rot.

Sentinel — detection

Runs every second, collecting latency, CPU, memory, GPU, queue depth and throughput. It uses z-score analysis against a rolling history of 20 samples to catch anomalies before they become failures. Every anomaly is categorised and severity-scored.

Raw metrics travel at medium priority. Detected anomalies escalate at high priority immediately — the bus does not treat "something is wrong" as ordinary traffic.

Sentinel is also the component that nearly killed the project. See the latency crisis.

Serpentine — classification

Synthesises everything Sentinel produces into one unified stress state across five levels: stable, moderate, critical, failure, catastrophic. It derives them from three indices — pressure, congestion and instability.

Catastrophic requires 92% pressure or 95% instability. Not lower. The thresholds are deliberately unkind because a classification that fires early is a classification nobody believes, and an escalation chain nobody believes is decoration.

Aegis — containment

Stops failures cascading. Nodes above a 20% error rate are isolated. Nodes above 95% load are throttled. Global circuit breakers activate under catastrophic state. Every action is logged with explicit reasoning attached.

Atlas — rebalancing

Redistributes load. Overloaded nodes above 80% shift work to underutilised nodes below 50%, targeting 65% after the shift. When redistribution is not enough, Atlas recommends autoscaler activation rather than pretending it solved the problem.

Punisher — repair

A dual-head neural network diagnoses incidents and ranks repair strategies by risk.

The interesting part is the governor rather than the network: an adaptive credit system, regenerating 0.02 credits per second to a maximum of 10, which prevents thrashing. As failures accumulate, credits deplete and the system becomes progressively more conservative. It cannot enter a loop of aggressive repairs, because aggression is a budget and the budget runs out.

Phoenix — recovery

Brings the system back after repair. Sequenced restarts in dependency order. Container rebuilds for nodes above 85% load. Global runtime state restoration.

Conservative by design: it performs more restarts than strictly necessary, never fewer. Restarting something that did not need it costs milliseconds. Not restarting something that did costs the recovery.

Hyperion — emergency shutdown

Fires within one second. Halts everything. No complex logic, no second-guessing, irreversible.

Hyperion is deliberately the least sophisticated component in the system. Anything clever here is another thing that can be wrong at the exact moment nothing else is working.

Terminus — the human kill switch

Terminus blocks orchestrators, AI agents and internal services from invoking it. It requires a human at a physical console with multi-factor authorisation. It encrypts a full system state archive before terminating.

Recovery afterwards is possible — but only deliberately, and only by a person.

That the system cannot reach the end of its own escalation chain is the most important property in this document. Everything above Terminus is the machine managing itself. Terminus is the point where it is not allowed to.