Epicenter

Break something

1 · Service
2 · Fault

One outage runs at a time on the shared lab; yours joins the queue.

Queue

How a live outage works

A real online shop of 10 services, with simulated shoppers browsing and buying; the charts above are its live telemetry.

  1. You pick a fault and it goes in.
  2. The alarm fires when the shop's telemetry stays abnormal.
  3. A Meta Muse agent investigates, never told what you broke. Its tool calls stream into the flight log.
  4. The verdict ranks five suspects, each with checked evidence, and is scored against your fault.

Or replay any of 468 recorded investigations, and see the numbers under Evidence.

Flight log

Verdict

The ranked causes appear here once the agent finishes.

No outage selected

Pick a recorded outage from the list. Each is a held-out benchmark case, replayed exactly as the agent investigated it.

Not sure where to start? This one uses every tool, metrics, logs, traces and the call graph, and finds a cause the best published method misses.

Watch an example

Flight log

Verdict

The agent's five ranked causes appear here at the end of the playback.

Play the investigation to unlock the known answer.

Find the root cause

You'll look at six recorded outages from microservice systems. Each shows the metrics that moved most around the moment an alarm fired. Pick the service you think caused the outage. For some outages you'll also see what Epicenter, an AI investigator, found. Work as you normally would; there's no trick and no penalty for being wrong. Time starts when an outage opens.