Direction 38: Time-travel debugging for Mo processes

The agent asks a running Mo system what any process was doing at any past point.

# usage pattern the agent programs against
snapshot = at(pid, time: T - 30.seconds)
state    = snapshot.state
message  = snapshot.in_flight_message
stack    = snapshot.call_stack
 
next_step = snapshot.step_forward   # replay one message
until_bad = snapshot.step_until { |s| s.state.balance < 0 }

Mo is unusually close to this already. Mo.Sim is deterministic; the failure model (chapter 3) is designed around replay from snapshot plus message log; fault injection is seed-reproducible. What’s missing is periodic snapshots at runtime and a query interface.

The bet: time-travel debugging is not exotic — it’s the natural output of features Mo already has. Not exposing it wastes them.

Cost is real: 4–6 weeks in the VM (see d36-vm-first-runtime), and needs a story for snapshot storage (in-memory ring, on-disk, both).