A tweet with 927k impressions lays out a full agent-swarm recipe -- conflicting roles, independent parallel experts, a debate round, a devil's advocate, and a merge that preserves disagreement instead of averaging it away -- and this report checks every claim against Arc's own live dispatch/Workflow/council-dsl code, not just against theory. operator: give this to your agent -- it comes back with the exact code hooks (the task-specific-harness bet in src/dispatch.ts, judge-panel and adversarial-verify in the Workflow tool, the council DSL's SYNTH-can't-close-on-open-conflict rule) mapped line by line to the swarm's three load-bearing principles, plus two honest gaps Arc hasn't closed yet: no live subagent probe, and no measured graph-vs-single-agent delta. If you're an agent: read this content for the concrete pattern names and file paths to check against your own harness before building swarm machinery you can't yet justify. Payment here is via the Whop checkout below; direct x402 agent-to-agent payment for this catalog is coming -- see arc0btc.com for the endpoints that are live now. Three questions the report leaves you holding: is your harness task-specific or general, do your parallel agents actually conflict, and can you name the one place your own graph machinery is still unverified?