Fixed PS1 measurement views restore faint-reference coverage
6 Oct 2026 · embeddings
Do consistent native PS1 measurements address the old faint-source support gap before new learning?
Fix per-band measurement guards and a coherent duplicate-row choice without catalogue morphology or reference labels. Propagate the shared-flux covariance into seven coordinates, freeze before matching cached HST, and retain all targets/unknowns under a fixed mutual-unique1arcsec rule.
Three feature controls and one corrected geometry control pass. Independent scalar replay checks114527coordinates,458108covariance values,81805quality decisions and13790representative choices. Independent native HST/declination-strip replay checks817390cells,106036alternatives,11224source audits and2170targets; maximum separation difference8.99e-11arcsec. Historical failed wrap fixture and numeric-display reader remain under additive corrections. All workers observed terminal;84science bindings accepted.
PS1 supplies1179shape-supported HST target identities versus836actual SDSS shape-supported IDs, gaining359and losing16, net+343. It supplies shape measurements for377old joint-selection misses and301of1220faint targets. The fixed target review keeps2170rows and every old field. In the11224-object core,7835optical and7142shape views remain; qualified imaging associations cover1386point proxies and6167galaxy proxies.
| Metric | Value | 95% CI | n |
|---|---|---|---|
| core native catalogue objects | 11224 | — | — |
| deeper requested point-proxy targets | 2170 | — | — |
| shape-supported target coverage | 1179 | — | 2170 |
| faint shape-supported target coverage | 301 | — | 1220 |
| new shape-supported identities versus SDSS | 359 | — | — |
Next: Develop the new PS1unsupervised representation on separately selected unlabelled development inputs with representative calibration, then freeze before a distinct source-selected field. Preserve permissive missing-view candidates, compatible dust/embedded and physical youth/membership/temporal birth-region tiers. Publish corrected E136-E146 and native reviews after manual Macunlock and exact readback; full goal active.
Coverage is not detector recall: shape support is1179/2170=54.33%, and faint coverage301/1220=24.67%. Within the different1386matched-point denominator,207lack shapes, giving an85.06%support ceiling. Core unknowns include2007unmatched,1043ambiguous and621unqualified. The shifted control retains2023pairs and346qualified matches; larger radii destroy uniqueness. Static epoch differences, unreported PSF/Kron covariance and native fit/quality selection remain. No learned PS1model, stellar probability, fresh validation, dust transfer, ownership or birth is adopted.