Model techniques map
Techniquesinference & servingagentic scaffolding

implementation detail · filed under inference & serving

Sandbox state forking

Forking creates a sandbox from the original’s exact state while leaving the original running, enabling side-effect-free reward judging.

Also called Fork.

source
1
model
1
lab adopt it
1
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

core 1

Documented in

Evidence

1 span quoted from the sources, strongest treatment first.

fork creates a new sandbox from the exact state of the original one while keeping the original running, which is useful for reward judging without side effects.

corepost trainingin AgentENVMoonshot AI

Filed alongside

Other methods under inference & serving :: agentic scaffolding.