Local Agent Handoff

Splits the follow-up work of Phase 1 among six roles, bridging the stopping rules defined in Document 05. Agent A (Sage Environment Builder) first sets up the Sage version, LMFDB release, database connection, and PARI/mwrank state, starting from conductor<150; Agent B (Algorithm 1 Reproducer) outputs the row count after each filter (initial → semistable/optimal/composite conductor → a3 → p-isogeny → ramification → rank/L-value → CLZ branch → Zhai branch → BSD(E,2)), saving the pass/fail reason for each curve; Agent C (2-Descent Referee) specifically monitors check_BSD_at_2—confirming the analytic Sha valuation, saving the bounds of each backend, verifying sha_an_ord_2 != 0 -> False, checking if timeouts cause false negatives, and explicitly forbidding treating $\dim\Sha[2]$ as $\operatorname{ord}_2\#\Sha$; Agent D (Algorithm 2 Cross-Checker) runs both the official Sage code and the pure-Python mirror of this package simultaneously, categorizing discrepancies into six types (number-field index issue, Kronecker convention, negative twist convention, finite-field point count, source-branch mismatch, official code drift); Agent E (Paper/Code Version Auditor) pins the arXiv version, GitHub file SHA, LMFDB release, and runtime flags, outputting the semantic diff between the paper's pseudocode and the current codebase; Agent F (Global Enclosure Referee) asks only three questions per round—did this round expand theorem coverage? Did it increase certificate strength? Or did it merely increase the enumeration volume?—if it is only enumeration volume, stop after three consecutive rounds and directly execute the stopping rules from Document 05.