Scan stage

The Scan stage (#scan) shows what the simulation will put around your agent. Your digital twin shows which of the systems your agent calls are real and which are mocked. Your knowledge base, when your agent answers from documents, shows those documents and the questions planted in them. Check both before you read any results. A wrong mock or a wrong document makes every result after it wrong too. The stage fills in when you scan your agent. The knowledge base is added when the suite build creates it. The stage tab counts what it found, for example 4 mocked.

​

Your digital twin

What the Agent really reaches, and the mocked surfaces standing in for the rest, so every scenario stays controlled and free of side effects. The figure shows your agent wired the way the harness runs it. Each pin is one surface your agent reaches: a tool, a skill, a sub-agent or a retrieval source. Point at a pin to read why it sits where it does. The section header counts what it found, for example 4 mocked or 6 surfaces, and lists the kinds it wired: 3 tools, 1 sub-agent.

​

The surface table

Open a pin, or use the Lens control to switch between live, mocked and live probes, to read each surface in detail:

​

What to check

• Every surface that writes to production, sends messages or moves money is under Mock environment. A surface under Your systems is called for real on every conversation.
• The Seeded world matches the records your agent expects, such as accounts, orders or tickets.
• If the agent answers from documents, Grounding reads Answers from your documents and Your knowledge base appears below.

To change what is mocked or seeded, say so in chat, or run /edit-env. See Build the suite and Ground in real records. In a conversation, each tool call says who answered it: served by the twin or your system. See Run results.

​

Your knowledge base

This section appears below Your digital twin when your agent retrieves from documents and a knowledge base was built for it. Link to it with #knowledge. Every document a spine, its passages hung in reading order under their section titles and indented by section. The header counts the documents and passages, for example 3 documents · 214 passages. If a file could not be read, it says so, for example 1 file not ingested. Point at a passage to see where it sits and what asks about it. Select it to read it. The passage view shows its Section, the text, and Asked in this Suite: each planted question, the scenario, and the turn that asks it. Use the arrow keys to move between passages. The Suite stage header tells you how much of the knowledge base the suite tests, and links back here with See it on Scan. See the coverage line. To fix a wrong fact or a bad question, say so in chat or run /edit-env. See Use a knowledge base.

​

Next steps