10 venues × 20 pairs × 5 order sizes ·
order-book snapshot every 5 min, averaged over the last 12 hours 192 rounds · 422,400 observations · 08-26 15:00 → 02:59 KST ·
measured from Osaka, Japan
Each figure is the median across all rounds in the window —
slippage in bps against mid, lower is better.
A cell is dropped from the ranking if it has fewer than 100 readings
or fails to fill more than 5% of the time.
Rank over timehow the standing moved, and whether it was us or them
VenueSegmentOrder size
Median slippage (bps; lower is better)
Only one snapshot so far — the trend appears once
the collector has run a few more times.
Columns sorted best → worst for the selected size ·
rows by market cap · cell colour = rank within that pair
Rank in row:bestworst|
hatched unfilled — the book could not fill that size ·
faded n=… — too few readings yet (a venue we only started reaching recently);
both are left out of the ranking|GRVThover a cell for detail
Spikes — how often it blows out, and how far. Median hides this
SCOPEBaseline is that cell's own median — comparing raw bps would make
BTC look calm and ADA look wild no matter what. Lower is better.
Slippage over time
How to read this. Every venue is lined up at its own normal level, and what is
plotted is how many bps worse than normal a reading came in. Same unit for every
venue and every pair, so one chart holds all of them. Left–right = how often it happens. Up–down = how bad the worst 1% gets.
Bottom-left is a venue you can plan around; top-right is one that surprises you.
★We do not use "× normal" as the headline. GRVT's BTC median is 0.006 bps, so a
2.5 bps reading is 398× — alarming as a ratio, minor as a cost.
Measuring the gap in bps removes that distortion.
Dispersion — bar = p10–p90, box = p25–p75, light line = median. Wider means it depends on when you trade
PAIR
Listing coverage — pairs a venue does not list are absent, not zero
How far to trust these numbers
Method — public order books were fetched and a market order of the given USD size was walked
against the book; the gap between average fill price and mid is reported in bps. Buy and sell were
measured separately and pooled. Contract units — OKX, KuCoin and MEXC quote size in contracts, so contract size was
applied. Without it they would look 100×, 1000× and 10000× deeper (after conversion, BTC
top-of-book lands at $117k–$4.75M across venues — the same order of magnitude). Price band — fills stop at ±10% from mid, which exchanges reject anyway. Without this
cap a deep book gets swept to the end and slippage looks better than it is. Unfilled orders were never invented — when the book ran out, slippage was left
empty and the cell is marked unfilled. Padding the remainder at the last price would make thin
books look liquid. Not yet modelled — tick/step rounding and minimum notional, quotes that vanish
before execution, DEX sequencer latency, USDT/USD depeg (Extended quotes in USD). Timing — captured 19:37–23:38 KST, i.e. before the US cash session
opens. RWA rankings must be re-measured during US market hours. Fees excluded — tiers and promotions differ by venue and would blur the ranking.
Add taker fees on top for real cost.
SIZECOSTMARKET
source 13 files / recent 12h · aggregated by _agg_liquidity.py · single offline file
Ranking uses only cells that filled the full size (≥ 95%) with n ≥ 100 observations.