Independent referee completed locally: HOLDS; NEW confirmed: yes under the job's printed-number rule. See VERDICT.md (under 40 lines). From this directory, reproduce with: nice -n 10 env OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 .venv/bin/python audit.py nice -n 10 env OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1 .venv/bin/python plot_audit.py Independent code reconstructs rate profiles from the primary Table S2 PDF and original released sequence CSV; only candidate/default coordinates and row-label manifest are read from producer outputs. No producer/referee program is imported or run. Float PAVA agrees with SciPy, and a separate 384-bit Arb computation verifies the candidate stress and PAVA conditions. The published figure's vector marks are evaluated separately with axis scales from its printed tick labels. Sources and output hashes are recorded in results/independent_audit.json. Live source checks and their limitations are in PRIOR_ART.md. Original analysis code would remove the remaining inference about implementation identity. No external contact or remote upload occurred.