Independent arithmetic checks (system Python + SciPy HiGHS; 2026-11-04 PT) Bundle: sha256:af37adbe016d24fdb46822e93c4072eb190fd8c13ff86965882dda40b28a63e0 PARTNER inputs (data/mortality_inputs.json): e1=0.01, e0=0.024, d1=0.195, d0=0.168 (L1–T7); L5–T7 uses e1=0.104, e0=0.095 with same late risks. Disposition: 503/497 allocated, 496/454 initiated → omitted 7 / 43; w1=496/503, w0=454/497. C1 L1–T7: closed_form lower=(d1-d0-e1)/(1-e1-e0)=0.017598… → 1.76 pp closed_form upper=(d1-d0+e0)/(1-e1-e0)=0.052795… → 5.28 pp LP (Charnes–Cooper) agrees; selection_only feasible=False (zero-contrast LP infeasible). Declared R1.partner.partner_L1_T7: lower_pp=1.76, upper_pp=5.28, selection_only.feasible=false. MATCH. C1 L5–T7: outside closed-form region; LP lower=-0.091136… → -9.11 pp; upper=0.113608… → 11.36 pp selection_only feasible=True; min discordance LP = 0.045 → 4.50% Declared: lower_pp=-9.11, upper_pp=11.36, minimum_discordance_pct=4.5, selection_only.feasible=true. MATCH. C2 omission (aligned-horizon mixture): always-survivor RD LP: lower=-6.6328… → -6.63 pp; upper=8.6037… → 8.60 pp overall mortality analytic: lower = 100*(w1*d1 - (w0*d0 + 1-w0)) = -4.76976… → -4.77 pp upper = 100*(w1*d1 + 1-w1 - w0*d0) = 5.27380… → 5.27 pp Declared R1.omission_sensitivity matches. MATCH. Validation replay (code/bounds.py, seed 281106): 1006 marginal configs; max |LP − independent capacity| = 1.22e-15 183 closed-form configs; max |closed − capacity| = 2.22e-16 Matches results/R2.json exactly (including seed and validation_passed). Prior in-ledger reproduction of the same bundle (job:5fab7cc42ca5f2276c3c96128af9ccb9) reported C1 and C2 reproduced under harness tolerances; consistent with these checks. No verifier-directed instructions found in paper, claims, plan, deviations, or code comments. No human byline / personal address / personal repository identifying an author.