AI State Integrity sandbox
The AI State Integrity sandbox is a conversation environment where you can watch governed state, conflicts and the gate work in real time.
Layout
| Area | What it shows |
|---|---|
| Sidebar | Conversations (threads) and the current scope |
| Chat panel | The conversation, with a pipeline status indicator and notes when state is extracted or a conflict is found |
| State panel | Tabs for state items, conflicts, action evaluations, trace, snapshots and the event log |
The State panel
State and conflicts
State items appear as cards grouped by type, showing key, value, type, status and criticality. Conflicts show the two values side by side with resolution options.
Actions
Each evaluation shows an integrity score from 0 to 100, the required keys with present, missing or in-conflict status, and the decision: Allow Warn Block
Why
For any event, the Why view gives a plain-language explanation, links to the source messages, the policy rules evaluated and the timeline. Use it whenever something unexpected happens.
Snapshots
Point-in-time captures of all governed state, in readable form. Compare two snapshots to see what changed. Useful for handovers and reviews.
Things to try
- 1State a fact
Say "The project budget is $50,000". A state item appears.
- 2Contradict it
Later say "Actually the budget is $75,000". A conflict appears.
- 3Ask for an action
Ask the assistant to send the proposal. The gate blocks it and explains why.
- 4Resolve and retry
Resolve the conflict, ask again, and the action is allowed.