Distinct regional optical and Gaia inputs; unchanged inference frozen before fresh accuracy

Can the unchanged unlabelled stellar representations be transferred to a distinct regional cohort before interpreting fresh reference labels?

Four predeclared disjoint directions, 32 optical plus four Gaia and four primary-spectral GETs. Preserve native observables, errors and epochs; no class/brightness query filters, fitting, threshold changes or winner selection.

Seventeen existing analytic controls. Independent 320540 optical,47192 Gaia,1856 spectral primitives; exact identities and native features/missingness. Independent87420 optical scores,116560 posteriors,18288 native Gaussian inferences and38862 supported Gaia embedding values. Maximum optical3.376e-13, Gaia ratio3.726e-9/chi-square7.451e-9. All workers observed exit0 and134 terminal bindings.

14570 optical sources,2776 Gaia sources,232 spectra; zero overlaps with74049/16668/2431 earlier identities. Optical9010 supported/5560 unknown; Gaia2286 supported/490 unknown. Frozen latent optical5422, density5271, raw3978, brightness482 provisional entries; Gaia learned1854/raw1851. Native response1.712MiB gzip for11.813MiB decoded. Prior stellar/dust/103molecular tiers retained.

Next: Evaluate these already frozen scores against the already acquired fresh spectra with the predeclared optical/Gaia geometry, full requested/unknown denominators, fieldwise burden, raw baselines, full curves and fixed review budgets. Retain the >=90%precision/>=80%recall gate and full robust stellar/dust/birth-region scope. E121 upload remains pending, with zero requests.

Inputs and inference only: fresh precision/recall, source associations and review-budget performance remain unevaluated. The two catalogue namespaces can overlap physically; do not sum as stars. Missing inputs, target-selected spectra and fixed regional selection limit inference. Candidate-count differences are not embedding superiority; no new discovery, physical class, youth, membership or birth.