If we are willing to accept that there are fundamental and enduring truths about traveler behavior that can be represented in mathematical equations (i.e., accept the premise of travel models), than it follows that the extent to which these truths are represented in our planning tools should be a useful indicator of how robust these tools’ predictions will be across time and space. This was, and remains, a key promise of activity-based travel models (ABMs): the behavioral fidelity of these models should allow for better predictions and increased transferability (across regions). These promises have been realized in the San Francisco Bay Area. Using an ABM with largely borrowed coefficients (hence the transferability) and calibrated to year 2000 conditions, we found re-calibrating (at the time sufficient observed data became available) the model to either 2005 or 2010 conditions unnecessary, as the validation statistics for these model years were, on whole, no worse than, and often superior to, the year 2000 statistics. Relative to 2000, the Bay Area in 2010 had 5.4 percent more residents, 13 percent fewer jobs, less traffic in many corridors, and lower ridership on most transit services. Taking the former two trends as inputs, our travel model accurately predicted the latter two trends. For example, the percent root mean square error of the morning commute highway assignment for all roadways (comparing observed to simulated traffic volumes) is 33 percent in 2000, 32 percent in 2005, and 25 percent in 2010. The error between daily observed and simulated transit boardings is plus 5 percent, plus 5 percent, and plus 2 percent in 2000, 2005, and 2010, respectively. This paper presents a variety of observed versus predicted comparisons across model years – to Census data, traffic counts, and transit ridership – and finds, in most cases, the travel model accurately predicts – without re-calibration and with largely borrowed coefficients – the prevailing trends.