Each lab is a fictional AI mission that goes wrong in a specific way. The scene shows you where to look, you zoom into the agents and calls that matter, then you change one thing and replay the same mission to see what it changes.
Interactive simulation Fictional missions, synthetic payloads and simulated metrics. Imperium captures the AI activity, policy decisions and enforcement evidence. Observatory reconstructs the mission from those events and explains the findings.
Featured investigation
Incident Model spend climbs while the quotation stops moving.
Research summarizes the customer’s documents. Then Pricing sends the same documents to the model again. Follow both calls, compare them side by side, and see what reusing Research’s results changes.
You can try Reuse research results
Run investigationPlayable investigations
Incident Model spend climbs while the quotation stops moving.
Research summarizes the customer’s documents. Then Pricing sends the same documents to the model again. Follow both calls, compare them side by side, and see what reusing Research’s results changes.
Run investigationIncident An inventory tool times out and the whole quotation waits.
Pricing retries the same inventory request while validation, the draft and the approval wait behind it. Follow each retry on the critical path, see what changed between attempts, and test a permitted fallback.
Run investigationIncident The policy says DENY. The tool returns the fields anyway.
A child agent asks the CRM for two restricted fields. Imperium records DENY, no enforcement follows, and the response carries both fields. Follow the decision, the missing enforcement and the response as three separate records, then enforce the denial at the tool boundary.
Run investigationIncident Every agent is green. The approver rejects the quotation.
Every agent finishes and validation passes, but the stock came from a day-old cache and nothing was reserved. The approver rejects the draft. See what validation checked and skipped, then require inventory confirmation before submission.
Run investigationOn the roadmap
