{
 "version":88,
 "analysis":"Exploratory fixed-spec household conditional association; not independent confirmation or causal identification",
 "origin_panels":[2018,2020,2021,2022,2023],
 "primary_domain":"v85 strict membership exactly reproduced by v87",
 "sensitivity_domain":"v87 outcome_roster_relaxed; event-year roster remains fixed",
 "outcome":"D = I(outcome RFOODS >= 2) - I(baseline RFOODS >= 2); both periods must pass v85 reported-path/official-recode validity rules",
 "exposure":"Any baseline household member's qualifying economic employer separation under frozen v85 event rules",
 "low_deposits":"Baseline checking/savings deposits < annual baseline household income / 12, frozen v85 construction",
 "within_cohort_model":"FINYR3 weighted least squares: D ~ intercept + economic + low_deposits + economic*low_deposits + baseline_food_insecure + (index_age-40)/10 + log(annual_income/10000) + baseline_household_size",
 "primary_estimand":"Equal arithmetic mean of the five adjusted within-cohort economic*low_deposits coefficients; units percentage points; positive means a larger separation-associated change in food insecurity among low-deposit households",
 "covariates":"Baseline only; no post-event income, deposits, household size or employment controls. No covariate search or post-result specification change.",
 "missing":"Complete prespecified covariates; report loss; do not impute or change index",
 "availability_gates":{"pooled_exposed_n":30,"each_cohort_four_cells_raw_n":5,"each_cohort_four_cells_kish_n":5,"all_240_replicate_designs_full_rank":true,"all_five_cohorts_required_for_equal_cohort_primary":true},
 "gate_scope":"These are exploratory household-model computational/sparsity gates, not the v83 holdout confirmatory gate. Passing does not establish adequate precision or overlap. Failed cohorts are reported unavailable, not silently dropped from the target.",
 "uncertainty":"240 matching longitudinal replicates per cohort, Fay factor 0.5; refit all covariates every replicate. No replicate-number alignment between endpoints. Equal-cohort independent variance is secondary. Primary normal interval uses SE envelope=sum(cohort_SE)/5, which upper-bounds the aggregate SE if marginal Fay variance estimates are valid; asymptotic normality and marginal variance consistency remain assumptions. It is not a jointly replicated survey CI.",
 "precision_target":"95% normal envelope half-width <= 5 pp, assessed after fit; do not declare the thesis established on nominal significance",
 "report":"All cohort availability/results, equal-cohort primary, independent-cohort diagnostic and full predeclared composition sensitivity, including null/reversed/unavailable results. No p-value selection.",
 "calibration":false,
 "causal":false,
 "known_before_spec":"All v85/v86 aggregate counts and person-level v82/v83 results; v87 readiness computation is running. No new household liquidity-hardship contrasts have been inspected. Repository checkpoint is not external preregistration."
}