Frozen unsupervised screening transfers to separate EGS sources

Does the EGS calibration setting retain stellar-proxy recovery on new source-selected inputs without retuning?

Freeze each five-method threshold at its E149 fixed 236-source calibration ranking before nine unfiltered regional GETs. Keep E147 model, E146 views and independent Gaia/HST reference rules unchanged; use a disjoint fixed rectangle and retain all unknown/requested states.

Nine GETs yield 3,423 PS1 objects/4,051 stack rows and 266 Gaia sources, with no calibration/training ID overlap. Native replay checks 571,234 PS1 cells, 4,522 Gaia cells and 1,012,614 cached HST primitives. Scalar/inverse replay checks 167,727 embedding values within 3.42e-13; independent geometry/readout checks 356 motion alternatives, 2,948 HST alternatives, 2,437 audits and 11,763 curve points. Original local reader/writer/verifier failures and six partial exports are retained under additive corrections; all workers terminal, 90 scientific bindings accepted.

The unchanged conditional threshold selects 238 candidates with 139 independent astrometric stellar proxies, zero qualified extended references and 99 unknowns. It recovers 139/152 requested proxies (91.45%), distinct from matched recall 139/149 (93.29%). Transferred brightness retrieves 133 proxies from 242 candidates; centred raw shape 67 from 194. At the same 122-source budget, conditional retrieves 114 versus brightness 97. The narrow 90%/80% conditional-proxy gate passes without new-field tuning; exact candidate/source/miss collections remain.

Next: Obtain representative negatives and unknown/faint/low-motion/crowded evidence, then broader geographically independent validation. Use explicit shared reader/writer contracts to prevent adapter errors in future stages. Maintain permissive stars, compatible dust/embedded classification and physical youth/membership/temporal birth-region refinement. Publish pending findings after Mac unlock and exact readback; full goal active, E121 zero, 16 MiB stage/5 GiB reserve.

Population purity remains unmeasured: 99/238 candidates are unknown, leaving a 58.40%–100% positive-proxy fraction even if proxy labels were correct. Reference selection misses faint, low-motion, crowded, binary and embedded populations. Nine of 152 requested targets lack shape support; association/epoch boundary denominators differ. The adjacent high-latitude EGS field and previously parsed HST archive limit independence/transfer scope. Full and brightness also pass the proxy gate. No physical star discovery, all-sky purity, dust transfer, membership or birth is established.