Seven properties of the catalogue could be lost by a one-line slip in the ETL with every validator
passing and the weekly job publishing the result. validateStars and validateExoplanets now refuse:
- fewer than 8 800 stars of V 6.5 or brighter (8 886 measured), or Rigel, Deneb or Alnilam gone;
with no magnitude handed to placementDistancePc, 7 378 are left;
- a host luminosity that is not positive, or fewer than 90 % of planets carrying one (6 036 of
6 354): st_lum is published as log10(L/L_sun), and stored unconverted 3 307 read <= 0 and
Proxima's -2.82;
- fewer than 40 colours marked as read off a temperature (54 measured);
- an archive-placed star more than 1 mas from its planets' published position carried from
J2015.5 to J2000 (6.6e-8 mas at most measured; not carried, 6 277.9);
- fewer than 300 HYG stars folded into Gaia keeping their more precise Hipparcos distance (384),
or Tarazed and Eta Leo off theirs;
- more than 5 archive stars numbered other than their name hashes to (1 measured).
These come from an interrupted earlier attempt left uncommitted in this worktree; its other half,
asymmetric archive distance errors that nothing in the ETL wrote, and a placementDistancePc test
that failed, was discarded (saved in the scratchpad as uncommitted-at-start.patch). The naked-eye
check is new, and runs before the Hipparcos one, which the same slip also trips.
Controls, each an ETL run from cache in a throwaway worktree that reached "Validating output..."
and failed with the named check's own message: no magnitude to placementDistancePc (naked-eye
floor); `10 ** logLuminosity` -> `logLuminosity`; colorFromTemperature never set; the archive
epoch offset times 0; combine never taking the other entry's distance; archive ids in arrival
order. The unmutated baseline passed. star.model.ts's count of flagged hosts, 57, is now 54.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
SenrokaiandClaude Opus 5.5
8d59c720b3