AI: Hyperbolic

I’ve spent much of this week revising the Tinderbox Hyperbolic View, which shows a map of links in a non-Euclidean space. Why do you need non-Euclidean geometry? Here’s a link map of part of my novel, Those Trojan Girls.

This is focused on one of the middle chapters, and we can see that some long roads lead here, and also that the outbound path to On The Road leads to repercussions.

The existing code was five years old, and my coding style has changed a bit. I never really trusted the geometry, because my grasp of the Poincaré disk was not really satisfactory. Once before, I’d tried a big refactoring to sort things out, but it had ended in a tangle.

This time, though, I had Claude and Gemini. The point was not to vibe code or to race; the point was that Claude does understand hyperbolic geometry better than I. So, I’d show Claude some dodgy code and ask, “Is this mixing up coordinate systems?” Too often, the answer was “Yes!” I’ve complained previously the Claude is sycophantic, but Sonnet 4.5 is more direct and will stand up to you when it is right.

So, in time we built separate classes for DiskPoint and ScreenPoint, and sorted out when to use what. Lots of methods got moved to the Point classes, and the class that calculates the curved lines between notes had to be completely revamped. (In hyperbolic geometry, a circular arc is, in fact, the shortest distance between two points.)

I’m not happy with the typography and I suspect the damping is too high, but it's really behaving far better than before.