Markets produce outliers; robust systems catch them. Pyth Network runs an always-on analytical shield to deliver clean, trustworthy prices. Its detection stack blends statistics, pattern checks, and incentives to keep erroneous or malicious inputs from influencing final aggregates.
Core logic evaluates entire distributions, not just pairwise comparisons. Techniques such as z-scores, interquartile ranges, and temporal consistency highlight suspect submissions. When flags appear, secondary checks—correlated assets, volume behavior, and a publisher’s historical accuracy—kick in, ensuring legitimate volatility isn’t suppressed while faulty data is filtered.
The framework learns as it runs. Thresholds adapt to market regimes, and economics enforce good behavior: publishers aligned with consensus accrue reputation and influence; frequent outliers face OIS penalties. Because aggregation is transparent, the community can audit and improve the system in public. The result is a reliability standard that rivals or exceeds many traditional vendors—purpose-built for open finance.
