An observer must call "signal or no signal?" from one noisy glimpse. Signal detection theory draws two bell curves on an axis of internal evidence: noise alone, and signal + noise shifted to the right. A single glimpse is one draw from one of them.
The observer sets a criterion — say YES above it, NO below. Slide it left and you catch more signals (more hits) but also cry wolf more (more false alarms). Slide it right and the reverse. That line is pure bias; it never changes how good you are.
How good you are is d′ — how far apart the curves sit. Pull them apart and every criterion does better; the ROC curve bows toward the top-left corner. The area under it is your sensitivity, independent of where you stand.
Two knobs, not one. Radiologists, jurors, memory, radar operators, lie detectors — all live on this bench, and confusing "cautious" with "skilled" is the classic mistake it cures.
The simulation stopped unexpectedly — the lesson continues without it. You can move on; nothing you did was wrong.