Measuring astrometric residuals against a fixed MPCORB orbit rather than a fitted one turns up residuals that a fit would have swallowed. Over 489 objects and 1.9M residuals, the W68 (ATLAS Chile) 2020-2026 Gaia2 bin has a clipped sigma of 0.150" and a raw RMS of 2390" — two-thirds of a degree.
Clipping keeps that out of the quoted sigma, so it does no harm to the weighting work, but a residual that large is not astrometric scatter. It is a mis-identification, a bad catalogue orbit, or an observation attributed to the wrong object — and it is invisible when the orbit is free to move, because the fit absorbs it.
Worth chasing for two reasons. If it is a linkage error it should be reported. And it suggests a cheap quality check that Layup could offer: predict from a catalogue orbit, difference against the observations, and flag anything beyond a few arcseconds. That costs a predict rather than a fit, and it is the sort of thing a survey pipeline would want before submitting astrometry.
Reproduce with ~/Dropbox/claude_layup/weights/fixed_orbit.py; the residuals are in per_object_fixed/*.npz keyed by station, year and catalogue.
Measuring astrometric residuals against a fixed MPCORB orbit rather than a fitted one turns up residuals that a fit would have swallowed. Over 489 objects and 1.9M residuals, the W68 (ATLAS Chile) 2020-2026 Gaia2 bin has a clipped sigma of 0.150" and a raw RMS of 2390" — two-thirds of a degree.
Clipping keeps that out of the quoted sigma, so it does no harm to the weighting work, but a residual that large is not astrometric scatter. It is a mis-identification, a bad catalogue orbit, or an observation attributed to the wrong object — and it is invisible when the orbit is free to move, because the fit absorbs it.
Worth chasing for two reasons. If it is a linkage error it should be reported. And it suggests a cheap quality check that Layup could offer: predict from a catalogue orbit, difference against the observations, and flag anything beyond a few arcseconds. That costs a
predictrather than a fit, and it is the sort of thing a survey pipeline would want before submitting astrometry.Reproduce with
~/Dropbox/claude_layup/weights/fixed_orbit.py; the residuals are inper_object_fixed/*.npzkeyed by station, year and catalogue.