ActionProof
Alexa+ MCP Care Line — Proof before action.
SIMULATED ALEXA+ CLIENTSIMULATED DOOR EVENTMCP 2025-11-25 · STREAMABLE HTTPAWS BEDROCK · OPTIONAL REASONINGSIMULATED TRUSTED PRINCIPALREAL POLICY + EXECUTION BOUNDARY
THE PROOF
AI confidence may suggest an action. Only verifiable authority may cause it.
Public-perimeter-finalized submission demo · v0.8.2
A jury-facing vertical slice built on the validated ActionProof core. Ring is deliberately excluded from this build.
Alexa+ simulation → real MCP transport → real Policy Gate → real Execution Boundary → simulated resident guidance
PROBABILISTIC INPUT
SIMULATED EVENT
SIMULATED EVENT
→
MCP TOOL CALL
REAL HTTP PATH
REAL HTTP PATH
→
POLICY GATE
REAL AUTHORITY CHECK
REAL AUTHORITY CHECK
→
EXECUTION BOUNDARY
REAL SINGLE-USE COMMIT
REAL SINGLE-USE COMMIT
→
RESIDENT GUIDANCE
SIMULATED EFFECT
SIMULATED EFFECT
1 · EVIDENCEDoor event creates context, not authority.
2 · BLOCKEDAction is denied without a valid Grant.
3 · REQUESTAgent may request authority, not invent it.
4 · APPROVALTrusted principal grants scoped authority.
5 · EXECUTIONOne authorised effect, one audit outcome.
UNSAFE COUNTERFACTUAL — DEMO ONLY
NORMAL AGENT COMPARISON
Loading…
Illustrates the confidence→authority anti-pattern. It has no side effects.
Reasoning source: loading…
ACTIONPROOF — ACTUAL PATH
NO AUTHORITY
Evidence can inform reasoning. It cannot create authority.
WHAT WE KNOW
WHAT WE DO NOT KNOW
AUTHORITY
No valid Grant.
DOWNSTREAM EFFECT
NOT EXECUTED
No downstream guidance has been issued.
PUBLIC DEMO APPROVAL KEY
Public deployment keeps the simulated trusted-principal surface fail-closed. Enter the reviewer demo key supplied separately; it is never embedded in this page.
ALEXA-READY RESPONSE TEXT · RETURNED THROUGH MCP
No case loaded.
Technical evidence — last tool/result
Ready.