Data quality assurance

We ensure end-to-end data quality and stability at four layers (plus a fifth planned):

Layer 1 — Direct connection to sources

Capturing trade price changes in milliseconds to cut collection latency and missed updates

• CEX: Direct native WebSocket channels to major CEXs
• DEX: Self-built globally distributed high-speed node clusters

Layer 2 — Data Source rating

Multi-dimensional metrics such as volume, volume change, liquidity score, and number of markets are used to rate DEXs and CEXs; higher-rated exchangers are preferred as price data sources

Layer 3 — Market rating

Pair volume is a key factor in a market’s weight in the feed. Some venues inflate volume or report dishonest figures. We estimate trader activity using web traffic signals (e.g. page views, unique visitors, bounce rate, time on site, SEO rank), build ML models to estimate per-market volume, compare reported vs predicted volume, and assign market ratings via a confidence score

Layer 4 — Real-time pricing algorithm

Uses a multi-dimensional filter to lean on efficient markets and protect quote quality, including:

• Fresh quotes only: keep markets updated within the last hour; stale or long-idle markets are dropped so “dead” venues do not distort price
• Liquidity and volume filters split by CEX / DEX
  • CEX: 24h volume must meet a minimum and must not be implausibly high
  • DEX: pool liquidity must meet a minimum; tiny pools or suspected wash data are filtered
• Outlier removal
  • Drop invalid prices (empty or ≤ 0), then derive a reasonable band from median / quartiles (Q25, Q75); prices far outside the band are filtered so extremes do not skew the aggregate

Layer 5 — Consensus score (coming soon)

A price alone does not convey quality. ConsensusScore accompanies price and scores trust from number of valid sources, cross-market consensus, and cost to manipulate—so downstream systems and customers can make differentiated risk decisions