debias() silently no-ops on obs80 catalog codes (single-char codes vs MPC_CATALOGS names)

[matthewholman]

Summary

debias() silently applies no correction when the star-catalog is given as an obs80 single-character code (e.g. W, L, q) rather than a catalog name (e.g. Gaia3, 2MASS, UCAC4). Since Obs80DataReader supplies the single-char code, debiasing never fires on obs80 input — including the bulk MPC observation archive.

Root cause

MPC_CATALOGS maps name -> code:

MPC_CATALOGS = {"USNOA1": "a", ..., "UCAC4": "q", ..., "Gaia3": "W"}

and bias_dict (from generate_bias_dict) is keyed by the codes (MPC_CATALOGS.values()). But debias() looks the catalog up as a key:

catalog_key = MPC_CATALOGS.get(catalog)   # name -> code
if catalog_key is None or catalog_key not in bias_dict:
    return ra, dec

So a name works ("UCAC4" -> "q" -> in bias_dict), but a code ("q") is not a key -> .get returns None -> the guard returns the astrometry unchanged. No debiasing.

Evidence

Fitting the same objects with debias=False vs debias=True gives byte-identical elements, because every obs80 catalog code no-ops:

    obj    da(deb=0)    da(deb=1)  changed?
  20000    3.829e-04    3.829e-04        no
  90377    4.078e-02    4.078e-02        no
    944    5.271e-08    5.271e-08        no

The obs80 astCat values (L, N, S, U, W, b, c, d, o, q, r, u, v, w) all match MPC_CATALOGS values, none match its keys.

Impact

The full-catalogue run ingests obs80, so the "faithful to JPL" weighting+debiasing story is incomplete: weighting applies, debiasing does not. (It does not move the headline element agreement much — debias is a sub-arcsec correction and well-observed objects are already tight — but it should actually run, and matters for bias-sensitive objects.)

Proposed fix

Accept a code as well as a name — bias_dict is already code-keyed, so let a code pass through:

catalog_key = MPC_CATALOGS.get(catalog, catalog)   # name -> code, or a code passes through
if catalog_key is None or catalog_key not in bias_dict:
    return ra, dec

Names still map to their code; obs80 codes now resolve directly; blank/None/unknown still fall through the not in bias_dict guard unchanged (preserving #401). Companion to #401.

Found while enabling debiasing on the MPC full-catalogue fit.