451a3621

By: ValentinHirschi <valentin.hirschi@gmail.com>

record GL638 runtime rescue and sampling comparisons

Document the 54-case A/P assessment, three matched 20-core 30k pilots,
20 current-binary H/Z controls and ten maximum replays. Preserve compact
inputs, complete numerical records and portable runtime presets. Clarify
historical z-probe false acceptance versus current Euler rescue, exact
coincident residues, and the density needed at the remaining hard corner.

All numerical panels completed with the frozen public binary and unchanged
read-only full-UV state. Structured artifacts, preset equivalence, hashes
and local links were checked; independent agents audited the interpretation.

User prompt:
Yeah I don't exactly remember what these two modes `relative_spherical` and `spherical_product_common_radial`, i think it's just normal spherical cartesian produce but with another overall hyperradius scaling, but I don't think it's useful to you anyway.
And yes please push your fix in the hyperspherical jacobian if you're sure about it too.

So what kind of sampling is needed to tame that remaining integrable singularity in GL638? And is arb prec not enough for your stability issues in GL638; can you explain the issue better there?

Timeline