In PR #129 we tested the improved routines for wavelength/LSF fitting on our original selection of arc lines across different channels, with no significant improvements.

As an experiment in the NIR channel, we matched (as much as possible) DESI arc lines selection. This (combined with fitting a higher polynomial degree) lead to an improvement of ~0.03 (from 0.07 to 0.04) Angstroms in some cases. We can match DESI's selection in all remaining channels. This should significantly improve b channel wavelength residuals.

Related to this, the following are the remaining bugs in wavelength fittings in different cameras:

Clear outlying fibers in b3:

Systematic offset in some fibers (possibly block boundaries) in z3:

Clear break close to the middle of the chip (fiber ~331) while fitting a 5th order polynomial in z1:

The last one we explored a bit more. We tried replacing lines that show this bimodal behavior in the residuals with no success. This feature is present in all calibration epochs only in z1.