A defect sink that formats every emitted capsule as a Logger line.

Not started by default — per decisions/2026-09-04-defect-reports-are-a-shipped-feature.md, mob is never the collector, and a sink that runs unbidden would violate that. An app opts in explicitly:

Mob.Defect.Sinks.Dev.start_link()

or under its own supervision tree:

children = [
  # ... app children ...
  Mob.Defect.Sinks.Dev
]

It is Dev because Logger output is the medium the connected-agent workflow already uses — mix mob.connect streams the device's Logger output into the agent's IEx session, so a defect logged here reaches the agent for free. An app that wants a production sink (a remote endpoint, a file, a push channel, or nothing) writes their own subscriber; the API is Mob.Defect.Bus.subscribe/0 plus a handle_info({:mob_defect, capsule}, ...).

What it logs

• severity: :fatal | :critical → Logger.error

• severity: :warning → Logger.warning
• anything else → Logger.info

The severity mapping is what routes a real bug to the log level a triager reads first. Logger.info for everything would bury a fatal alongside a perf regression; Logger.error for everything would train the reader to ignore the channel — exactly what the redaction and dedup work goes into preventing.