The refactor is back on track...

[Daniel Lee]

I finally found out what's causing numerical issues. Turns out that starting from a value that's not identical to the old one (down to the last bit) causes the results to be fairly different.

On the unconstrained space, the old one initialized the one unconstrained parameter to:

1.5641332541198749

The buggy one initialized it to:

1.5641332541198751

That was enough to send everything off the rails.

I was getting that due to initializing on the unconstrained space, creating a var_context on the constrained space using transforms, then letting the inverse transform get it back to the unconstrained space.

I eliminated that step, so we're good again.

That took way too long to find.

Daniel