SFS bottleneck audit: executed audit under the A2 error model (Release B2, unrefereed candidate)

Executed identifiability audit under the frozen A2 linkage-aware error model (doi:10.5281/zenodo.21893667). 4,963 rows; 4,884 bounded with 50-digit dual certificates; 79 class rejections; zero aborts. Certified family-union findings at the tight complexity budget: both independent YRI processings — the bottleneck claimants' own spectrum and the Cousins–Durvasula processing — read depressed, not severe under the most information-preserving declared setting: the ancient-window depression ratio is certified within [0.162, 0.934] and [0.132, 0.961] respectively across the entire twelve-mapping claim-window family, certifiably below the recent baseline and certifiably far above FitCoal-level severity. The out-of-Africa contrasts (CEU, CHB) certifiably exclude severity and never certify depression, matching the dispute's shared premise of an African-specific signal. Across the full concession ladder the identified sets widen to uninformative: every certified statement is conditional on its declared error-model setting, and the severity question remains class-identified. Status: anonymous-workflow, unrefereed, producer-run. Archives the exact bytes of GitHub tag v0.2.0-candidate (sha256 7f1232ffff70bcc1b8616d31170aeac87240ccf5f8d7f78aa167a844b526d566).