We report a 20-year campaign to track the 1.8 hour photometric wave in the recurrent nova T Pyxidis, using the global telescope network of the Center for Backyard Astrophysics. During 1996-2011, that wave was highly stable in amplitude and waveform, resembling the orbital wave commonly seen in supersoft binaries. The period, however, was found to increase on a timescale P/P-dot=3x10^5 years. This suggests a mass transfer rate in quiescence of ~10^-7 M_sol/yr, in substantial agreement with the accretion rate based on the star's luminosity. This is ~2000x greater than is typical for cataclysmic variables of that orbital period. During the post-eruption quiescence (2012-2016), the star continued on its merry but mysterious way - similar luminosity, similar P/P-dot (2.4x10^5 years). The orbital signal became vanishingly weak (<0.003 mag) near maximum light of the 2011 eruption. By day 170 of the eruption, near V=11, the orbital signal reappeared with an amplitude of 0.005 mag. It then gradually strengthened to its normal 0.08 mag amplitude, as the star declined to its "quiescent" magnitude of 15.7. During the ~1 year of invisibility and low amplitude, the orbital signal had increased in period by 0.0054(7)%. This is probably a measure of the mass ejected in the nova outburst. For a plausible choice of binary parameters, that mass is at least 3x10^-5 M_sol, and probably more. This represents >300 years of accretion at the pre-outburst rate, but the time between outbursts was only 45 years. Thus the erupting white dwarf seems to have ejected at least 6x more mass than it accreted. If this eruption is typical, the white dwarf must be eroding, rather than growing, in mass. Unless the present series of eruptions is a short-lived episode, the binary dynamics appear to be a mutual suicide pact between the eroding white dwarf and the low-mass secondary, excited ... . (etc., abstract continues)