AI systems generate intentions, recommendations, and signals. Operational execution causes real-world state changes. Parmana sits between these two layers — enforcing admissibility, evaluating policy, and issuing bounded execution authority before any consequential action proceeds. This separation is the foundational design principle. An AI system that governs itself cannot provide the independent auditability that regulated environments require. Governance must be structurally separate from the intelligence it governs.Documentation Index
Fetch the complete documentation index at: https://docs.manthan.systems/llms.txt
Use this file to discover all available pages before exploring further.
The separation model
Governed signals flow from AI systems into the admissibility layer. Policy evaluation determines whether those signals meet the criteria for execution authority. Only after authority is issued does operational execution proceed — and every step is recorded in an append-only lineage that supports replay reconstruction and portable verification.AI Governance Separation
Why this positioning matters
| Layer | Responsibility |
|---|---|
| AI Systems | Generate signals and intentions |
| Governed Signals | Bounded, schema-validated inputs |
| Admissibility | Structural validation before policy evaluation |
| Policy Evaluation | Deterministic rule application |
| Execution Authority | Bounded permission to act |
| Lineage | Cryptographic record of every decision |
| Portable Verification | Independent auditability by any party |
See also
- Execution Authority - how authority is issued and consumed
- Governed Signals - how inputs are bounded by schema
- Portable Verification - how decisions are independently verifiable
- Deterministic Governance - why reproducibility is required