What is Karma? #
Karma is an experimental event-driven system for understanding change. It explores how normalized events, state, dependencies, and outcomes can support observability, analysis, and evidence-based decisions.
Adage defines the infrastructure control model. Karma provides the workload and proving ground. Desired-state configuration, Terraform components, and SSM discovery belong to Adage; Karma experiments with the observations and analysis that can demonstrate whether the model works in practice.
Source, prototype, and proposal #
| Capability | Evidence | Status |
|---|---|---|
| Event/action envelopes | Schemas and topic conventions | Defined in source; universal enforcement is not established. |
| CDC normalization | Kafka Streams implementation and Python source | Code and mappings exist; end-to-end integration requires verification. |
| State and latency analysis | ClickHouse SQL | Definitions exist; live processing is not certified here. |
| Graph API | Lambda handlers | Mock graph responses and logging stub; no persistent graph integration. |
| Prediction/deviation | Job source | Placeholder code and notes. |
| Action execution | Examples | Pseudocode placeholders; no verified action engine. |
| Neptune graph, CLI/service, coordinated changes | Theory | Design proposals, not established runtime capabilities. |
Intended event path #
Raw CDC → normalization → events.normalized → ClickHouse event history and state/latency views → analysis → proposed actions.
Normalizer and SQL source exist. Source/sink wiring must be supplied and tested; detection and execution stages are incomplete. The API prototype is not automatically connected to this streaming path.
Practical next steps #
Use the source README for component-specific setup. The Compose file starts supporting Kafka/ZooKeeper and ClickHouse services; it does not deploy a complete application. There is no root Poetry application or finished Karma CLI in this checkout.
First validate one synthetic event path and a queryable result. Separately complete the infrastructure cost proof. Larger graph and automation claims should follow observed evidence.