We resolve a long-standing mystery of how the transport model AMPT manages
to generate sufficiently high elliptic flow (v2) with only few-millibarn
2->2 partonic cross sections. With the help of the covariant Molnar's
Parton Cascade (MPC) transport solver, we investigate the influence of the
particular initial conditions, interactions, and dynamics encoded in the
partonic stage of AMPT. Through detailed comparisons, we pinpoint at last
which features enable AMPT to circumvent the "RHIC opacity puzzle" raised
by Molnar and Gyulassy [NPA697, 495 (2002)].