Think of building the flexibility matrix experimentally, one column at a time: to get column j, you apply a unit load only at coordinate j and measure how every coordinate on the structure moves.
So the flexibility matrix's \(j^{\text{th}}\) column comes from applying a unit force at coordinate j and reading displacements everywhere.
A third way to settle this is purely from the units of a flexibility coefficient. Each entry \(f_{ij}\) of the flexibility matrix has units of displacement per unit force (for example, mm/kN for a translational coordinate), since it physically represents a displacement produced by a unit applied force.
Only applying a unit force and measuring displacement gives a ratio with the correct displacement-per-force units of a flexibility coefficient.
Therefore, the correct answer is a unit force is applied at coordinate j and the displacements are calculated at all coordinates.