Explicit infrared hypotheses restore scoring coverage but not a learned ranking advantage

Can frozen unlabelled infrared families retain singleton and ambiguous inputs without forced ownership and improve young-source retrieval?

Before reference readout register native band/map anchors, all reliable cross-band companions within a full 5-arcsec diameter, an explicit no-companion view, a 256-view overflow guard, uniform mean/min/max scores and primary 10%/fixed14/fixed18 budgets. Retain a separately predeclared reliable 100-micron diagnostic and rejected-list extension. Reuse frozen E153 families; no fit or download.

Four geometry/ambiguity/overflow controls pass. Independent full Cartesian replay checks 190722 admissions, 34127 scalar inference/moment quantities within 4.264e-14, and all 717 native inputs. Independent raw-reference replay checks 305466 VISION primitives/95402 nulls, 36956 HOPS primitives, 5051982 geometry admissions, all 483 audits, 32/84 contexts and every class/cohort/common-colour/budget union within 8.995e-11 arcsec. Workers observed terminal; 39 scientific bindings accepted.

All 717 anchors have evidence across 1336 explicit hypotheses, 283 multi-view anchors and zero overflows. Primary 18 of 173 reliable core anchors recover 5/26 HOPS and 5/77 VISION, including 2/8 Class0, equal to brightness on HOPS. Eight overlapping youth-context anchors represent only five HOPS identities, with ten unknown anchors. Separate reliable 100-micron diagnostic: 18 reviews retrieve 13/26 HOPS, 13/77 VISION and 6/8 Class0; brightness retrieves the same 13 and six. Preserve both exact collections and all previous candidates.

Next: Remove brightness nuisance from spectral-family evidence and compare actual measured-band contrast/colour baselines; keep cheap 100-micron/singleton follow-up. Broaden unlabelled infrared/dust development and independent representative starless/contaminant calibration/transfer; improve shorter infrared, shape and ownership evidence. Preserve ordinary optical/Gaia stars and compatible cloud/youth plus physical membership/temporal birth-region scope. Full goal active; 12MiB stage/5GiB reserve; no downloads, images, E121 requests or Git rewrite.

No learned ranking advantage: mean/min/max/anchor-only/brightness each recover five HOPS at primary 18 reviews; brightness wins at 20%. Single-band family evidence can reflect brightness. Uniform hypotheses are heuristic, ranges are sensitivity, not calibrated ownership or youth confidence. Repeated band/map anchors are not distinct stars; context channels overlap. Reliable full-budget availability caps at 20/26 HOPS; all 483 reliable/rejected anchors reach 21/26, not a passed broad recall/purity gate. Targeted PACS exposures, upstream supervised quality, sparse consumed references and missing negative truth preclude population precision, physical classes, membership, ages or births.