Bistro Research
Partial

Compatible optical inputs cover all thirteen sparse or missing WISE stellar targets

5 Oct 2026 · embeddings

Imported record: produced before this platform existed
Question

Can a native optical view remove the measured input-coverage ceiling before further unlabelled stellar learning?

Hypothesis

Acquire64 unfiltered primary-photometry tiles for the eight consumed E113/E125 directions; keep native signed flux/error, no class features or new classifier. Six colour/shape coordinates with approximate shared-measurement covariance and complete unknown/ambiguity denominators.

Test

Five analytic controls. Independent732220 native primitives,219666 features,1317996 covariance entries and36611 metadata-poisoned sources; maximum covariance difference1.137e-13. Independent5056366 geometry decisions/646 alternatives,563 audits and22 missing-input review rows; maxangle2.976e-11arcsec. Additive quota/provenance export repairs preserve original failures.

Finding

36611 distinct native optical entries,22819supported/13792unknown;24.50MB decoded retained as3.80MiB gzip-only. All13 E125 missing/unscorable in-field stars gain supported unique optical alternatives. Optical supports20/21quality field targets; one different optical ambiguity retains WISE support, giving21/21 compatible input availability. E113quality field49/67opticallysupported. No model fitted or recall improvement claimed.

Next: Learn a label-free error/missingness-aware optical stellar embedding with calibration/test source exclusion, raw morphology/colour/brightness baselines and explicit stellar/QSO semantics. Conservatively retain optical/IR/dust/103molecular alternatives and full requested denominators. Freeze before a distinct survey cohort. Pending E121 upload remains untouched; full robust goal active.

Caveat · open

Input availability is not classification/detector recall. Consumed sparse spectra, primary/clean photometry population selection, static geometry and different survey selection functions. Optical ambiguity and E113coverage remain; no native image ownership/motion correction. Approximate independent-band/PSF-model covariance omits real correlations. No adopted union, physical class or birth.