Overview

By the end of this act you can: deploy a service, hit a failure you didn’t write, and work backwards from the symptom to the cause using the agent. This is the integration test for everything in Acts 1 and 2.

The arc

1. Deploy the same codebase you changed in Act 2.

2. Hit the failure. A latent bug, one you didn’t author, surfaces only once it’s running.

3. Troubleshoot from the symptom (a log line, a 500, a stack trace) back to the cause, using the agent.

Why this lands

Because you already oriented in this codebase in Act 2, Act 3 is genuinely about agent-driven debugging under pressure. You’re not re-comprehending everything from zero while also fighting a deploy failure.

TODO: gated on the codebase decision

The specific latent bug (config/env mismatch, missing permission, serialization difference, timeout) is chosen with the app and deploy target. Keep it separate and latent so it doesn’t contaminate Act 2.