igmhub / picca / 32980652959

62%

master: 63%

push

gihtub

fix(4most): make the readers work on the real delivery

Three problems surfaced on first contact with the data:

- The joined delivery carries case-variant duplicate columns ('specuid' and
  'SPECUID', 'snr' and 'SNR'), so renaming in place raised
  "Column SPECUID already exists". Build a fresh table instead.

- Forests were never rebinned onto Forest.log_lambda_grid. Every other reader
  calls forest.rebin() after construction; without it the rest-frame trimming
  never happens and the mean-continuum stack is empty.

- 4MOST stores flux in raw erg/(s cm2 Angstrom), so values are ~1e-17 and
  inverse variances ~1e34, while the continuum fit starts from a mean
  continuum of order unity. Minuit failed to converge on 77 of 79 forests.
  Scale to 10**-17 erg/(s cm2 Angstrom), as SDSS and DESI do, and declare it
  in 'flux units'.

Also skip spectra with no valid pixel rather than letting rebin raise.

End-to-end on the real delivery: 1549 spectra -> 812 objects -> 811 after
filtering -> 524 forests with a fitted continuum. Weighted delta mean
-0.0005, weighted std 0.54, observed-frame stack flat to |delta| < 0.005.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

1408 of 2881 branches covered (48.87%)

Branch coverage included in aggregate %.

6592 of 9965 relevant lines covered (66.15%)

1.98 hits per line