Population design-probability correction

Final publication audit identified a reporting defect in the original population ledger. All 223 selected rows/weights and the frozen stratum allocation are correct, but unsampled eligible rows have sampled_stratum_groups=0, numerator zero and blank inverse weight. These encode realization, not the design probability. Every eligible group in a stratum must have the same design inclusion fraction min(30,N)/N, whether selected or not. Excluded groups keep zero/zero and blank weight. review_sample alone records realized selection.

Preserve the original ledger, queue, summary, checker and publication records. Produce a separately named corrected population ledger by changing exactly sampled_stratum_groups, inclusion_numerator and inverse_inclusion_weight on unsampled eligible rows. No IDs, group, stratum, exclusion, selected case, outcome, threshold, score or measurement changes. The original queue remains byte-identical. The earlier independent checker mirrored the realization convention; its receipt does not establish the corrected population interpretation.

Freeze input hashes and this amendment before transformation. Separately verify every original field, uniform group/stratum design probability, selected queue identity and complete hashes. Worker/checker each <=60 seconds, one CPU, sampled RSS <=256 MiB, outputs <=2 MiB/1 MiB, reserve 5 GiB. No model or acquisition. Add corrected downloads to existing releases without overwriting original scientific bytes. Current Git/GCS release revisions remain immutable snapshots.