Unlabelled astrometric families recover optical gaps; raw motion matches labelled retrieval

Can noise-aware unlabelled astrometry extend stellar retrieval beyond missing optical counterparts and reduce contaminants?

Expand class-free exclusions to all direct spectral coordinates over1998-2026; eight families on5000unlabelled Gaia sources, noE128training. Freeze embeddings/native likelihood ratio before E113operating calibration; compare raw chi-square and unchanged optical, retain all unknown/requested targets.

Five corrected analytic controls; independent2565456interval decisions/646alternatives/611withheld and5000training roles. Native inverse/determinant109040Gaussian inferences/231710embedding values; maximum posterior1.027e-14/mean5.152e-14/covariance9.454e-13/ratio3.726e-9. All1238references,254calibration/830development curves,75budgets and10189Gaia/28394optical candidate records replay. Registered geometry/export/reader-path corrections preserve originals, terminal stdout seal amended separately.

145updates converge. Gaia E11366STAR/0other of67positive, E12517/0of22, E12863/0of66; raw chi-square same labelled identities. Whole requested quality field65/67,17/21,60/65unique Gaia versus49/67,20/21,60/65optical. Learned Gaia10169/16668candidates, raw10174, noncalibrated union10189. Optical refinement7006 while all28394parent records retained. Linked146existing spectral stellar proxies plus every stellar/dust/103molecular tier.

Next: Freeze before another distinct fresh cohort; evaluate requested targets, uncertainty, low significance/unknown retention and scalable review. Keep compatible optical/IR/dust/molecular tiers; improve only where learning demonstrates value versus raw baselines. Embedded/dust and birth-region refinement remain required; E121pending permission, full goal active.

Consumed labels are not fresh validation or all-sky candidate precision. E125recall77.27%fails80%; unknown score31/48/81in conditional Gaia cohorts, omitted associations remain requested. No learned advantage over raw on labelled references. Gaia5%review423finds6/66E128STAR, optical astrometry1872finds44/65versus parent49/65+2QSO. Gaia catalogue/spectral bias, sensitivity/epoch approximations and high candidate burden; no calibrated union, physical birth or new-star discovery.