Clarify AANA value proposition

README.md

@@ -21,14 +21,25 @@ This card describes AANA as a control-layer architecture and runtime package, no
 agent proposes -> AANA checks -> agent executes only if allowed
 ```
 
-## What AANA Provides
+## What AANA Adds
+
+Most existing approaches are prompt instructions, moderation/classification,
+LLM-as-judge review, framework middleware, or opaque provider-side model
+alignment. AANA is different because it standardizes the pre-action decision
+itself:
+
+```text
+agent proposes -> AANA checks -> tool executes only if route == accept
+```
+
+AANA adds:
 
 - A public Agent Action Contract v1 for pre-tool-call checks.
-- Python SDK and CLI helpers for local checks and audit-safe summaries.
-- TypeScript SDK helpers for JavaScript/TypeScript agent runtimes.
-- FastAPI service endpoints for HTTP integration.
-- Adapter families for privacy, grounded QA, agent tool-use, and cross-domain action checks.
-- Audit-safe decision metadata: route, AIx score, hard blockers, missing evidence, authorization state, and recovery suggestion.
+- Evidence/auth-aware routing across `accept`, `revise`, `retrieve`, `ask`, `defer`, and `refuse`.
+- Hard execution rules: wrapped tools do not execute unless AANA returns `accept` with no hard blockers or schema failures.
+- Correction and recovery suggestions for missing evidence, missing authorization, unsupported claims, or human-review escalation.
+- Audit-safe decision metadata: route, AIx score, hard blockers, missing evidence, authorization state, and redacted log event.
+- Python SDK, TypeScript SDK, CLI, FastAPI, MCP, and middleware surfaces with validated decision-shape parity.
 
 ## Public Boundary