An 8.6 percent sampling-correlation allowance permits a half response in 30 years in 54 network summaries at a fixed score limit

We test the effect of missing fast-slow error covariance on common lag-mixture bounds in 54 published network summaries. We retain the source slopes, rate grid, marginal score variances and common weights. At the prior fixed score limit, zero covariance gives a 30-year response range of 0.195 to 0.413 of the spatial prediction. Allowing all sampling-error correlations gives 0.012 to 0.630. A certified sampling-correlation magnitude allowance of about 0.086 first permits a half response. If the allowance is recentered four units above the new best profile score, the range is 0.138 to 0.387. Allowing all total residual correlations gives fixed-limit bounds of 0.009 to 0.666 and recentered profile bounds of 0.148 to 0.430. All 66 endpoints have attainable weights and global certificates. These are conditional score bounds, not confidence limits, ecological forecasts or estimates of the missing covariance. Common weights, fixed spatial slopes and marginal variances remain assumptions. Code, source hashes, proofs and aggregate results are included.