Local Multi-Agent Handoff Prompts

Splits Phase 1 among six roles: Agent A Proposition Auditor (builds the theorem dependency DAG for BSD-W/BSD-F/BSD-S using only standard external mathematics, forbids equating rank equality with full BSD, equating analytic Sha with actual Sha, or using Wikipedia as a core theorem source); Agent B Banwait–Huang Reproducer (reproduces the algorithm from arXiv:2601.16044, running the authors' samples first and then conductor≤500000; if any paper criterion has fields that cannot be exactly determined from LMFDB/Sage/Magma, mark as unknown, no guessing allowed); Agent C Certificate Schema Engineer (implements and validates bsd_curve_certificate.schema.json, explicitly separating numeric evidence, rigorous computation, external theorem, conditional theorem, and actual proof); Agent D Rank-2 Wall Analyst (centered on 389.a1, audits item by item how r_alg, r_an, Ω, Reg, c_p, Sha, L^(2)(1)/2! are calculated, whether they are rigorous, and whether they are supported by theorems); Agent E Adversarial Referee (looks for circular BSD assumptions, numerical-to-proof leaps, finite-to-global leaps, p-part-to-full leaps, etc., in all outputs; output must be PASS/FAIL/OPEN, not just general advice); Agent F Internal Theory Isolation (audits the relationship between Neo.K's old lattice/PRC drafts and BSD, only allowing outputs that can be translated into standard lemmas, circular or undefined parts, and required new proof obligations, forbidding bringing internal axioms directly into the external main proof). This six-role handoff structure is itself a real-world case of the TRP spirit—generation, adversarial review, and neutral adjudication executed separately, just sliced finer than TRP's three roles.