The inversion of frequencies to (a,e) has known breakages. Diagnostics suggest these are coming from bad frequency calculations when going forwards (from a,e to frequencies). Adding a break for invalid frequencies might help.
Failure modes have been fixed; there don't seem to be any issues beyond numerical precision. I'm closing this unless there is a clearer example of a failure mode.
The inversion of frequencies to (a,e) has known breakages. Diagnostics suggest these are coming from bad frequency calculations when going forwards (from a,e to frequencies). Adding a break for invalid frequencies might help.