Given a two-body scenario with explicit gravity-on/off and tangential-on/off triggers, the learner predicts the trajectory: gravity ON + tangential ON → orbit (curving around the central body); gravity ON + tangential OFF → falls straight in; gravity OFF + tangential ON → flies off in a straight line; gravity OFF + tangential OFF → stays at rest. The learner correctly handles the M5 anti-targets D2 (no-tangential → fall-in) and D3 (no-gravity → straight-line, NOT fall-in).