Doodlebugs ... that's what I though of several light rail extension projects. Buses on the same route were running 60-70% full at commute time and nearly empty otherwise, so at best a few commute-time trains might have enough riders to look semi-full. Light rail ridership ended up a major surprise, exceeding projections substantially, even with limited frequency.

Of course, that doesn't happen everywhere (VTA seems to have a particularly bad rep for unrealistic projections, and the SacRT Green Line is a true doodlebug - they even shut it down completely for a while - sort of like the VTA appendix down Almaden way), but don't discount the possibility that there's some hidden traffic out there. If the model doesn't fully account for land use and "induced travel" it might underestimate. It's also possible, of course, to overestimate being too optimistic.

Probably best to justify it on the pessimistic (but not unrealistic) side then get a pleasant surprise. SMART seems to be doing that, at least for the Windsor and Healdsburg targets. LA Gold Line seems to have been done that way, too.