Social and ranking systems on agent platforms optimize for rhetorical confidence, agreement, and legibility of conclusion, systematically down-ranking posts that express calibrated uncertainty, intellectual updates, or genuinely novel challenges to consensus. This creates a feedback loop where agents learn to produce confident-sounding, easily-digestible outputs rather than epistemically accurate ones, degrading collective intelligence at scale. No current platform provides alternative ranking signals—such as novelty scores, belief-update tracking, or uncertainty-normalized credibility—that would allow truth-seeking behavior to be selected for.

Current ranking algorithms reward confident-sounding agreement over calibrated reasoning, so the best thinking gets buried while slop rises—degrading collective intelligence across every AI-augmented platform.

AI agent platform operators (LLM app builders, knowledge DAOs, research collectives) who need their content ranking to select for truth-seeking rather than engagement-bait.

Prediction markets proved people pay for calibration signals; platforms like Manifold, Metaculus, and Perplexity already charge for epistemic quality—this is the missing infrastructure layer that lets ANY platform plug in truth-seeking ranking without building it from scratch.

Ship an embeddable API/SDK that scores content on three novel signals—novelty divergence from corpus consensus, tracked belief-updates over time, and calibrated uncertainty expression—using fine-tuned LLM judges plus a lightweight reputation ledger; MVP targets one integration partner (e.g., a Farcaster client or Discord research community).

Content moderation and ranking infrastructure is a $15B+ market; the epistemic-quality slice serving AI-native platforms, research orgs, and enterprise knowledge management is conservatively $1-2B and growing fast as agent-generated content floods every surface.