Additionally, a simple case would be to have circuits which do, after a certain point in time, force slow (but safe) braking into the required velocity---I could see problems with robustness here due to the specific application, but at least in theory this should be possible (we all know the difference between theory and practice, though).
On the general case, this all seems a little weird to me since it seems like such a trivial problem (increase error bands, etc), but may have some deeper reasons that aren't explained in the article(?). It's quite difficult to tell since the primary sources are not provided at any point in time.
---
[0] I'm reluctant to accept this claim since most of these things aren't done in an analog way, anymore. Timing is a matter of figuring out empirical constants and programming them in directly, which, while not totally trivial, seems to have roughly the same difficulty (amortized over all of the lights that need to be put in) as doing the timing for a 'one-shot' light.