Overview
CMI is built around one question: does the trading printed by a venue, on a given pair, appear to originate from genuine order flow?
It provides the per-venue, per-interval statistics needed to answer it, for any covered market, across every venue that trades it, on a common timestamp grid.
The idea behind every workflow
Comparing a market to its own past cannot separate a manipulated market from one reacting to real news. Comparing it to its peers at the same timestamp can. Every venue trading BTC-USDT at 14:32 UTC sees the same news, the same flow and the same price, so their trade statistics should look like siblings: different in size, similar in shape. When one venue's shape differs, the market did not change; the venue did.
Comparison at the same instant cancels the factors that make time-series baselines noisy, such as news, volatility regime and time of day. It is also harder to game, because fabricated statistics have to stay consistent with peer price action, and that costs money. CMI returns one row per venue per interval to make this comparison direct.
Contents
| Objective | Reference |
|---|---|
| Field definitions and semantics | Field Reference |
| Assess whether a venue's reported activity can be trusted | Volume Integrity Assessment |
| Determine whether a price gap was liquidity or deliberate pressure | Price Dislocation Analysis |
| Monitor whether a stablecoin is holding its peg | Stablecoin Depeg Detection |
The strongest signal
Distributional tests (digit patterns, size histograms, trade timing) measure the bookkeeping of trades, and bookkeeping is cheap to fabricate. Price impact measures the economic footprint: how far price moves for a given amount of trading. Making fabricated volume produce a realistic footprint requires actually moving the price and defending it against arbitrage, which shows up as losses. Price impact is not a returned field, but it is one line of arithmetic over high, low, volume and vwap.
Practical notes
History starts 2024-01-18. For 2020 through early 2024, use the MDA API reference.
Data is aggregated into 1-minute buckets, so sub-second patterns are not visible.