Control Plane

ClickHouse governs data exhaust.
Cosmocrat governs decision exhaust.

The control plane that enforces, records, and replays every AI and human decision—by policy, not convention.

Runs in your environment. Your keys. Your data.

What is decision exhaust?

Data exhaust captures what happened—logs, metrics, and traces of system events. Decision exhaust captures why it happened—the reasoning, policy evaluation, authority checks, and state transitions that led to an action.

Decision exhaust is the audit-grade record of every AI and human decision — including intent, policy evaluation, authority, execution, and outcome — produced at runtime.

For deterministic software, logs are sufficient. For probabilistic AI agents that reason and act autonomously, data exhaust is blind to intent. You cannot audit an AI's decision to refund a customer by looking at a database transaction log alone; you need the governance record of the decision itself.

Ungoverned
AI Agent
SQL Query
Payment
The Governance Gap

Agents don't just chat.
They execute.

Decision exhaust exists whenever an agent reasons, calls tools, touches data, or executes actions. Without a control plane, these distinct events are ungoverned, unrecorded, and risky.

"Experimental AI creates hallucinations; enterprise agents require audit trails."

The Architecture of Governance

A dedicated control plane for decisions, sitting alongside your data plane.

Business Apps
Agents
Workflows
Humans (CRM)
Data Exhaust

ClickHouse

Observability Speed Layer

Decision Exhaust
C

Cosmocrat

Decision Authority Plane

The Governance Cycle

A single runtime path for every action.

01

Policy

Evaluate authority, limits, and context before execution.

Fail-closed by default.
02

Enforce

Permit, block, retry, or escalate—across every tool and lane.

No silent failures.
03

Receipt

Produce an audit-grade record bound to inputs, policy, and outputs.

Cryptographically attributable.
04

Rollback + Replay

Reconstruct the full decision path and tighten policy from outcomes.

Continuous governance improvement.

Receipts you can audit (and replay)

Every governed action produces an execution receipt: who/what/why, what policy evaluated, what state transitions occurred, and what actually happened.

Data Exhaust: A Log Line
2026-01-18T05:12:44Z tool_call=refund_user amount=42.00 status=200 provider=stripe latency_ms=183
Useful for debugging. Not sufficient to prove authority.
Decision Exhaust: An Execution Receipt
StatusAPPROVED (fail-closed policy)
Policyfinance.refunds.l1 — limit: $50
ActorSupportBot-01 (authority: L1)
Actionrefund_user(user_id=U-1872, amount=$42)
Outcomestripe.refunds.create → ok (id=rf_9K2...)
Receipthash=9d7b... • replayable
Receipts make governance provable—not just observable.
Receipts establish intent, authority, and outcome—independent of system state.

Frequently asked questions

Data exhaust is the byproduct of system operation (logs, traces). Decision exhaust is the byproduct of governance (reasoning, policy checks, authority grants).

One captures what happened. The other captures why it was allowed.

No. Observability watches the system. Decision exhaust is the system's legal record. It provides auditability and replayability, not just visibility.
Logs are mutable, unstructured, and reactive. Decision exhaust is immutable, structured, and causal—it links the intent to the outcome cryptographically.
Cosmocrat is designed for high-throughput runtime environments. Decision exhaust is captured asynchronously where possible, adding minimal overhead to the execution path.

Related Platform Concepts