Frozen Orion optical screen misses embedded young-source candidates
6 Oct 2026 · embeddings
Does the frozen unsupervised optical screen retain recovery in a crowded star-forming region, including infrared-selected young sources?
Apply the unchanged E147 model/E146 native views and all five E150 thresholds to a fixed Orion rectangle before nine unfiltered regional GETs. Preserve native Gaia and cached revised VISION/HOPS class, censoring, duplicate and missing states; no new fits or threshold changes.
Native replay verifies358916PS1cells/2822Gaia primitives,305466full VISION primitives and36956full HOPS primitives. Independent scalar/noise replay checks19838coordinates/79352covariance entries/2721representatives and133329embedding values within1.04e-12; Gaia308motion alternatives match within2.77e-11arcsec. Independent static/reference replay matches1929audits,58/77/26requested targets and all classes/IR states/budgets within3.90e-11arcsec. Original IRS/export-scope checker failure remains under a documented additive v2 correction; all workers terminal,86science bindings accepted.
Core1929PS1objects have1186optical and574shape-supported views. Fixed conditional screening selects59candidates:41Gaia proxies,10VISION and1HOPS context, overlapping into43unique supported entries. Requested Gaia recovery is41/58=70.69%, VISION10/77=12.99%, HOPS1/26=3.85%. All84VISION and32HOPS core contexts and13-band measurement/upper-limit/nondetection states remain visible. Twenty of22HOPS optical-shape misses have measured70-160micron evidence;18have350/870micron support.
| Metric | Value | 95% CI | n |
|---|---|---|---|
| Orion provisional optical candidates | 59 | — | — |
| requested Gaia proxy recovery | 41 | — | 58 |
| requested VISION young-source recovery | 10 | — | 77 |
| requested HOPS young-source recovery | 1 | — | 26 |
| HOPS Class0 recovery | 0 | — | 8 |
| confirmed physical births | 0 | — | — |
Next: Register compatible unlabelled regional infrared/dust representation development and independent survey-source recovery, then representative starless-cloud/contaminant truth, calibration and broader transfer. Preserve optical/missing/low-motion tiers and physical youth/membership/temporal birth-region scope. Full goal active; publish pending E136-E151 findings and exact reviews after manual Mac unlock/readback; E121zero,12MiBstage/5GiBreserve.
The crowded/embedded transfer fails robust young-source recovery: all8HOPS Class0 and9VISION Class0 targets are unselected, HOPS flat0/11, VISION flat0/14. Optical shapes exist for only29.76%ofcore objects and4/26HOPS targets; loosening a threshold cannot recover absent inputs. Brightness retrieves44Gaia/17VISION/4HOPS targets at192reviews versus59conditional candidates; learning is not uniformly better. No broad negative truth makes population precision null. Reference channels/catalogues overlap; static heterogeneous epochs, crowding and selected catalogue coverage do not establish physical class, emission ownership, membership, age or birth.