By-origin battery design adjudication

Adjudication needed

The 25% candidate run and an adversarial feasibility audit (2026-08-26) surfaced design questions in the by-origin battery that need owner rulings — distinct from genuine model drift, which is being root-caused separately:

1. Max-of-five quantile envelope false-block rate. Simulating the envelope test against the fixed ASEC empirical distribution at each leg's realized effective support, six person legs fail more often than they pass even under a faithful null: qualified BDC income (24% pass), farm income (30%), W-2 wages from qualified business (36%), Form 4952 (39%), casualty loss (44%), negative estate income (48%). This is not purely a low-count problem — farm income has amount-leg ESS ≈ 1,480 and still 30%, because discrete mass near a tested quantile destabilizes a max-of-five rule.
2. Authority per leg. Some references are themselves imputed (e.g. alimony_expense: ASEC does not measure alimony paid; the ASEC-side reference is PUF-imputed — the comparison grades imputation against imputation). Others are cross-source clashes (alimony_income: survey-measured ASEC 0.056% vs PUF-fitted draws — the live incumbent's own asec/puf channels disagree 2.6× on the same leg). The battery currently treats ASEC-origin as truth for all legs.
3. Family-level error control. 135 correlated comparisons with per-comparison fatal thresholds; no multiplicity treatment.
4. What the incidence data does NOT support: a generic rare-leg support floor. At 25% support, no failing incidence leg is a luck problem (lowest ideal-null pass probability ≈ 86%); 19 of 26 failing incidence intervals sit wholly outside the band. Incidence failures are systematic and should be root-caused, not floored.

Companion issues cover the two deterministic clone-0 QBI comparisons and the weeks-unemployed infeasible constraint. Ruling should fix the test's design integrity without tuning anything toward passing a specific artifact.

Start by reviewing the by-origin battery and its max-of-five envelope, authority-per-leg, and family-level error-control logic, then compare the incidence findings with the companion QBI and weeks-unemployed issues. Done means an owner ruling resolves the four listed design questions and preserves design integrity without tuning toward a specific artifact.