Build for the unanswered question.

Implementation notes for developers who need an agent to tell the difference between a tool call and an observed result.

• A GitHub create call timed out. What now?

  How a contract and correlation marker help an agent recover a reference without blindly creating another issue.

• No match is not proof that nothing happened.

  Why incomplete evidence must stay indeterminate, even when an agent wants a yes-or-no answer.

• A tool receipt and an observed result answer different questions.

  Where receiving-system verification fits alongside traces, durable runtimes and connector tools.