Yes, it's finished, and yes, it works. Indeed, it works rather nicely, if I do say so myself, with a very satisfying snap action. But the $64,000 question is: how well do the trains run on it? Particularly, when compared to Eishindo's original version? It's time for the acid test.
I'll be perfectly honest: I wish I could deliver a definitive answer. The truth of the matter is, it's only somewhat better than Eishindo's. Of course, the fact that it's functional at all is a significant achievement! But it's not what I would term a gloriously overwhelming success, by any means.
The testing process went on considerably longer than expected, and actually started off as a gloriously overwhelming failure: mine appeared to be worse than Eishindo's. That embarrassment faded as I carefully studied what was going on. There were a great many variables to take into account—starting with the fact that, in my excitement, I'd neglected to clean the track.
The first big issue was derailments. This was certainly as perplexing as it was disturbing, since tracking was supposed to have been substantially improved by the new design. And the train I was using for the test (the Hankyu 9000) should have fared better.
Almost but not entirely ready to abandon the project altogether, I found that my 9000s needed a little fine-tuning: the trucks on the powered units were a bit snug, and barely able to swivel. Some extremely careful stretching of the springs improved matters. Even then, one of them simply refused to make it through either switch without derailing.
But the unpowered units between them wouldn't stay on the rails, either. The fact that they derailed about equally on the two switches, and only on the diverging route, suggested that perhaps the unpowered cars are simply too light. Or, perhaps the powered unit passing over the switch is stalling briefly, causing the cars in between to buckle. Whatever the case, I decided to continue testing without the unpowered cars for a while, and concentrate on just getting something to successfully make it through.
The problem of derailing gave way to one of stalling, which was every bit as puzzling. It's been a known issue with Eishindo's switch—which is not a surprise given its relatively huge electrical gap. But with no such gap on mine, why were the trains stalling anyway? Especially since a test meter indicated everything was live?
What I discovered was that there was some shorting between the backs of the wheels and the point rails on my switch. This wasn't a huge worry, as I'd actually anticipated it might happen (and it wouldn't occur had the points been electrically isolated, as I'd wanted). It was cured with a coat of paint on the outsides of the points.
Now I was able to see what was going on when the train wasn't rocketing through the switch and into the air—whereupon I encountered more surprises. The first thing I noticed was that points were actually shifting a little as the powered units rolled over them—and only on the straight route. This was a real head-scratcher!
It turns out that the point magnet for this route was much weaker than its counterpart, and this allowed the magnetic wheels to pull the points askew—just enough to cause the wheels to drop. That's when I learned that there was considerable variability in the strength of the point magnets (which could constitute a QC issue, in the unlikely event this design went into production). With a few spare switches on hand, I was able to swap out the wimpy magnet with a more energetic one, and this solved the shifting point problem.
This then led me to the next discovery: the gears on the powered trucks are almost exactly the same diameter as the wheels, and as a consequence, they could occasionally come into contact with the point rail for the opposing route. This would either cause the car to jump, or the points to shift, or both. The solution was to taper the ends of the point rails downward so that the gears wouldn't pick at them. (And the challenge for me was to do this on the finished prototype switch without dismantling it!)
Very gradually things were improving, but not enough to satisfy me, and I began to wonder if the wheel flanges were striking the plastic fulcrum I glued in between the points. And so, in a fit of desperation, I removed the plastic part. Lo and behold, some of the derailment problems magically disappeared.
Yet I was still experiencing more derailments than I preferred, and finally I found yet another rather odd little clue: the non-powered trucks on the powered units were able to tilt very slightly back and forth. Because the springs cause the trucks to favor one position over the other, the cars actually "tiptoed" down the track on only two wheels of the lead trucks!
Since the corresponding truck on the Kiha was not like this, I concluded that it was a design fault, and I felt a little better knowing this was not my fault. I then set about modifying the tilted truck so that it would remain flat, just like every other truck on the whole train. After dismantling a 9000 and scrutinizing the problem, I learned it was easily corrected by simply trimming a small amount off of the truck's bolster pin.
It was, like all of the rest, a tiny incremental improvement. Unfortunately, up to this point, I was still testing only the powered units. With all of the changes that had been made so far, I got brave enough to return the entire train to the track. Alas, the non-powered cars still derailed—and still on both switches.
Ultimately I determined that it was not solely a fault of either switch per se; in studying the action in slow motion, I was able to see the cars leave the rails before reaching the frogs or points. I believe the problem is a combination of the couplers being too stiff and the switch geometry, which constitutes a tight S-turn. The former I might be able to address, but the latter was entirely out of my hands.
It was easy enough to test the track geometry aspect of my hypothesis: I simply replaced the switch with an S-turn track section, which has the same geometry as the diverging route of a switch. And the train proceeded to derail in exactly the same manner and corresponding location as it had on both switches. That was a relief—if still a disappointment. So I omitted the non-powered car that did the most derailing, and things improved considerably; indeed, performance of the three-car tests were quite impressive by comparison.
Perhaps the most vexing issue of all was that one powered unit occasionally derailed on the switch through the straight route. After studying that problem six ways to Sunday, I began to suspect it might be the point magnets fighting with the magnetized wheels: I spotted the truck doing a little micro-leap into the air just a fraction of a millimeter before coming into contact with the points. If true, this is a nearly impossible situation to resolve, as neither the switch nor the train could function without their magnets.
I was exhausted. After many hours of tinkering over the course of several evenings, my switch was finally performing at least a little better than Eishindo's. But it was never "perfect," as I'd hoped it would be. I concluded I might have to accept that there's no such thing as a "perfect" switch, particularly as there's no such thing as a "perfect" train to run through it. So I settled for a final report card grade of a B, versus Eishindo's C.
Could I ever achieve an A? I will at least try. This was, after all, the first of its kind; I learned a great deal making it, and I think I can make a better one (and I will still send one to Mr. Hirai as a gift). If I get really ambitious, I might even see if I could conjure up a longer switch, which ought to cure a number of lingering ills. We shall see!