1[2 {"id":"C1","text":"11 of 27 (procedure, cell) pairs are failure regimes (coverage < 0.90): CR0 in all G<=10 cells, CR1 at G=5 with ICC 0.2 and 0.5, and WCB in all G=5 cells; no procedure fails at G=30.","results_keys":["failure_cells","n_failure_cells","coverage"],"status":"supported","caveat":"cr0_G10_icc0.0 (0.8965) is only ~0.5 MC SE below 0.90."},3 {"id":"C2","text":"CR0 substantially underestimates Var(ybar) with few clusters (E[V]/Var 0.48-0.66 at G=5) and its coverage is 0.72-0.78 at G=5.","results_keys":["rel_bias_var","exact_rel_bias_var_analytic","coverage"],"status":"supported","caveat":"MC bias estimates agree with analytic values within 2.5 MC SE."},4 {"id":"C3","text":"CR1 has the highest average coverage over the G=5 cells (0.8905 vs 0.8527 WCB and 0.7395 CR0) but remains below nominal there.","results_keys":["best_procedure_small_G","avg_coverage_small_G","coverage"],"status":"supported","caveat":"Specific to Rademacher WCB with linear-interpolation quantile; G=5 Rademacher bootstrap is highly discrete."},5 {"id":"C4","text":"WCB gives the coverage closest to nominal at G=10 with ICC>0 and at G=30, and is consistent with nominal in all three G=30 cells.","results_keys":["coverage","nominal_cells"],"status":"supported","caveat":"Differences between WCB and CR1 at G=30, ICC=0 are within MC error."},6 {"id":"C5","text":"Only five pairs are consistent with nominal coverage (|p-0.95| <= 2 MC SE): cr1 at G=10 and G=30 with ICC=0, and wcb in all G=30 cells.","results_keys":["nominal_cells","mc_se_coverage"],"status":"supported","caveat":"No multiplicity adjustment across 27 comparisons."},7 {"id":"C6","text":"Monte Carlo SEs of coverage are 0.005-0.010 with R=2000.","results_keys":["mc_se_coverage","R"],"status":"supported","caveat":""},8 {"id":"C7","text":"The naive iid interval (baseline) covers ~0.95 at ICC=0 but only 0.29-0.49 when ICC>0.","results_keys":["baseline_iid_coverage","baseline_iid_rel_bias_var"],"status":"supported","caveat":"Supplementary baseline, not a required estimand."}9]10