Bind pose
The reference stance — often a T- or A-pose — a model holds at the moment it is bound to its skeleton, and against which all later animation is measured.
When the rig is set up the engine records each bone's placement in this pose. Its inverse, the "inverse bind matrix", moves a vertex from the model's rest coordinates into the bone's own coordinate system; from there the bone's current pose transforms the vertex back to where it belongs. Without this step vertex skinning would be impossible, because nothing would know where the skin "really" sits.
Modellers choose a wide, open pose deliberately — arms and legs spread reduce the chance that weights "bleed" between nearby body parts. A bad bind pose only shows up once the shoulder folds wrong in motion.