Doomclock launches five-model geopolitical forecast engine

This is a strong example of why multi-model systems need behavioral guardrails, not just more models.

• –Hiding prior probabilities reduced anchoring and made analyst outputs more responsive to new evidence.
• –Named prompt rules became shortcut tokens, so replacing labels with evidence-first questions produced better reasoning.
• –Search grounding improved source provenance but still allowed fabricated facts, forcing deterministic API injection for hard numbers.
• –The Black Swan cross-theater view is a practical product layer that turns raw model disagreement into decision-relevant monitoring.
• –Early traction is niche but credible for a technical audience (Project post with active discussion).