Early Virtual Science: some lessons for the AVO

Experience with ASTROVIRTEL, scientific analysis of current large data sets, and detailed preparation for the truly huge future missions especially GAIA, provide important lessons for the Astrophysical Virtual Observatory. They demonstrate that the science cases are impressive, specifically allowing new thresholds to be crossed. The AVO is more than just faster cheaper better, it allows the new. The example of use of pre-explosion imaging of supernova to identify progenitors is used to illustrate some general challenges. Some non-trivial technical astronomical issues arise, especially astrometry, to complement the many technical implementation challenges. A critical scientific lesson is the need to quantify data quality. How are we to ensure the Virtual Observatory produces top science, and avoids being overwhelmed with mediocre data?