Independent Referee / Local Agent Handoff

The wrap-up document for the v0.3 series, with an extremely clear goal: do not search for more curves; first try to overthrow the 696.e1 family theorem. It designs six independent referees, each targeting a different link in the argument, with a posture of "finding loopholes" rather than "confirming": Referee A re-verifies item by item the base conditions of Theorem 2.14 and the splitting conditions for symbolic $q\in\mathcal P$, outputting PASS/FAIL; Referee B re-verifies the odd ordinary/additive branches using only the non-semistable alternatives explicitly permitted by Banwait–Huang Remark 2.10; Referee C prioritizes using the stronger but simpler sufficient hypotheses of FW Theorem 1.1 rather than rewriting the most general Theorem 1.7 themselves, avoiding "grading one's own homework"; Referee D proves that the period does not divide the Manin-period discrepancy, and explicitly requires prioritizing published Manin-constant results, not relying on the unpublished Edixhoven remark; Referee E independently recalculates the entire Chebotarev computation from scratch (discriminant, Galois group, resolvent, fiber product degree, conjugacy class size, density); Referee F writes a program to scan all $\mathcal P$ primes for $q<10^7$ to find counterexamples, but the document explicitly reminds itself: a numerical sweep is not a theorem proof, it is only for finding implementation/case bugs. Finally, the document gives a clear stopping rule: if Referees A-E all PASS, upgrade to DERIVED THEOREM / PREPRINT CANDIDATE; if any item FAILs, freeze expansion and revert to that exact failed lemma.