polymathlisted

# Polymath Split-Brain Workflow A meta-method for mapping the possibility space of a problem before committing to an approach. Forces uncorrelated views by spawning parallel agents that each see the problem through a different disciplinary lens. The value comes from the **spread** of returned ideas, not from averaging them -- disagreement is the signal. ## What You Get - Landscape map: N sub-agent outputs covering uncorrelated disciplinary lenses - Synthesis of disagreements in structured template (not averaged consensus) - Actionable divergence points highlighting where frames conflict ## Killer items -- silent failures if skipped These are load-bearing. Skipping any produces output that _looks_ successful but is wrong: - **[K1] Step 0 triage produced an explicit go-decision** -- without this, the workflow burns lenses on a problem that didn't need them. - **[K2] Step 2 rosters were built by THREE PARALLEL sub-agents (close/mid/far tiers), each running max-distance walk within its tier WITHOUT looking at the problem's specific features** -- without this, orchestrator bias leaks into tier construction and the composed polymath personas lose their spread. - **[K3] Step 7 pre-spawn checklist was filled out and announced to the user before any sub-agent call** -- without this, the user has no chance to correct the framing or refuse the cost. - **[K4] All sub-agent dispatches went out in a SINGLE message with multiple tool blocks** -- sequential dispatch silently wastes 10x