Caltrain's 2021 peak hour timetable achieves a score of only 96, meaning the service is slightly worse than the taktulator's baseline, the 2011 peak hour timetable, which a decade ago earned our reference score of 100.
Why so mediocre?
It's mostly in the padding. Caltrain service planners have evidently given up on trying to run a tight timetable while also dealing with the debilitating variability in station dwell times inflicted by the lack of level boarding. The right way to solve this problem was and remains to plan for and implement level boarding, a system upgrade that (on a per dollar basis) has even greater service benefits than electrification. The lazy way is what we see here: about 20% of extra padding is baked into the station-to-station run times, allowing a train to easily make up time and arrive "on time" in the event of a dwell time delay along the way. In the absence of a delay, trains can dawdle along even more slowly than the ancient diesel fleet could manage, and just sit at stations until the clock says it's time to go.
Back in 2005, a baby bullet express with five intermediate stops was timetabled at 59 minutes. In 2015, it was up to 61 minutes. Today, the same express runs in 66 minutes. This follows a pattern noted by Alon Levy on the deterioration of speed.
Another factor that explains the lower score is one fewer train per peak hour, resulting in longer intervals between trains. This helps with fleet size, where only 16 train sets (+2 spares) are needed to operate the 2021 timetable where 18 (+2) were needed before. Two more train sets are freed up for maintenance downtime, a vital bit of breathing room as Caltrain's older diesel fleet is breaking down more often. The fleet is well past its expiration date due to the multi-year delays in the electrification project.
Is it optimal?
Can a timetable be devised that uses no more than 16 (+2) train sets and scores better than 96? Why yes it can. Here is a Silicon Valley Express timetable that uses just 14 (+2) train sets with four trains per peak hour per direction, and scores 102.
Why is it better? First, it follows census patterns and puts the stops where they link the most residents and jobs, not where there is the most parking. It is more regular and has fewer gaps with long waits. While this does not figure into the service score, it makes far better utilization of the train fleet (83% of the time in revenue service, versus 70%). More efficient fleet utilization leads to fewer trains and fewer crews being needed to provide the same service, reducing labor and maintenance costs per passenger mile. All this is done without any magic: 15-minute equipment turns, comfy 40-second station dwells and a slightly less absurd padding level of 15%. There are zero overtakes, so fewer opportunities for cascading delays. Finally, this timetable is much simpler to understand for a rider, having just two service patterns.
One can only hope that despite this interim state of mediocrity, Caltrain will successfully implement its "moderate growth" service vision, which scores an impressive 240. Getting there will require reliable 30-second dwells for which level boarding is a must.
Credit to Richard Mlynarik who did the time-consuming part of this analysis.