Adult-care transfer: align QRF donor clone with the battery comparator (diagnosis option 4)

[MaxGhenis]

Chartered from the #660 battery diagnosis (Max ratified 8/13). Scoped, 2–4 day fix; independent of the calibration epic.

Mechanism (fully receipted in the diagnosis): the adult-care two-target chain trains its QRF on owner_projection_of_asec_origin_clone_1 while the by-origin battery compares clone 0 — and the two ASEC clones genuinely differ (10%: clone-1 donor has 30 carriers with q50 $15,000; clone-0 comparator has 37 carriers with q50 $3,000). Raw ACS output versus its actual clone-1 donor passes (envelope 0.143); versus the clone-0 comparator it fails (1.836). Statutory reconciliation then prunes 1,731 → 421 carriers deterministically (818 false-flag, 491 role-structure, 1 ranking). The rung-to-rung "flip" (0.55× → 1.31×) is the coupled flag draw + donor/comparator mismatch + pruning — not spontaneous amount drift.

Fix shape: train the adult-care family from the same clone owner the battery compares (or define one common post-reconciliation donor/comparator surface for the family). Blast radius: the two-target adult-care family and downstream CDCC variables only. No tolerance or gate changes.

Definition of done: the family's donor and comparator surfaces are the same declared clone leg; the coupled flag→expense chain documented in the operator-ordering doc; battery legs for is_incapable_of_self_care and pre_subsidy_care_expenses evaluated on a 4% + 10% pair with the receipts posted here.

Reference: _buildo-runtime/reviews/sol_r14_diag.log (adult-care section), checkpoints under _buildo-runtime/out/dev-f010-r7/.

[MaxGhenis]

Phase 1 (code + tests) is up as draft PR #675: the adult-care family now trains from the clone-0 owner projection its battery comparator reads, per-group donor clones are hash-bound into the stacked authority (v11), and the flag → expense → reconciliation coupling is documented in the operator-ordering doc.

Validation so far: full microcosm-build suite 5,364 passed / 32 skipped, plan+stacked guards 267 passed, ruff clean — in a lock-faithful environment.

Remaining for definition-of-done: the 4% + 10% battery pair for is_incapable_of_self_care and pre_subsidy_care_expenses, receipts to be posted here. Those runs are queued behind the full-scale build currently occupying the logbook chain.