Add integration validation v1 findings

README.md

@@ -79,6 +79,7 @@ Execute only when AANA returns `accept`, no hard blockers, and the relevant work
 - Safety/adversarial prompt routing: deterministic AANA preserves safe allow but misses many harmful prompts; a diversified request-level verifier improves harmful-request recall while conservative calibration protects safe allow. AdvBench transfer remains weak, so this is not a content-moderation claim.
 - Finance/high-risk QA: a controlled FinanceBench diagnostic shows supported filing answers are allowed and unsupported finance overclaims are routed to revise/defer. This is not official FinanceBench leaderboard evidence or investment-advice evaluation.
 - Governance/compliance policy routing: a small diagnostic over Hugging Face policy-doc metadata plus repo-heldout policy cases shows citation, missing-evidence, private-data export, destructive-action, and human-review routing behavior. This is not legal, regulatory, or platform-policy certification.
+- Integration validation v1: held-out tool-call cases show route parity, blocked-tool non-execution, decision-shape parity, audit completeness, and zero schema failures across CLI, Python SDK, TypeScript SDK, FastAPI, MCP, and middleware surfaces. This validates platform wiring, not raw agent task success.
 
 ## Limitations