How we measure Hyperliquid apps on-chain
We make DarkTrade.
This page explains how the competitor figures on our comparison pages are made, from public files anyone can download.
Every app that sends orders to Hyperliquid with a builder code leaves each fill in a public file Hyperliquid publishes per builder and per day (stats-data.hyperliquid.xyz). We count an address as an app's only when DefiLlama's open registry of Hyperliquid builders lists it under that app (factory/hyperliquid.ts, commit f9003fd), and we say so: the attribution is DefiLlama's, not the app's.
What we compute
From those files we compute, over 7 and 30 full UTC days: the fee each fill carried in basis points (builder fee divided by the fill's value), how many fills came from trigger orders (stops and take-profits together), and how many distinct accounts traded through the address. Accounts are accounts, not people.
- Fills: every row in the window's files, on every market. A lower bound.
- Fee its fills carried: the builder fee divided by the fill's value (price × size), in basis points, over perpetual-futures fills on Hyperliquid's main market and its HIP-3 markets: the median per fill and the volume-weighted average. Spot fills are left out, because a spot buy pays its fee in the token it buys.
- Accounts: the distinct accounts that carried the code (never the other side of the trade). Accounts, not people. A lower bound.
- Fills from trigger orders: the fills flagged as coming from a trigger order, out of all fills, and out of the closing fills (the fills that reduced a position).
- Windows: the last 7 and 30 full UTC days before the measurement.
Why every count is a lower bound
The published files cover only part of each day, so fills and accounts are lower bounds: we print 'at least' and never scale them up. Rates are measured over the published part. We checked a sample against Hyperliquid's own order records: all 51 flagged fills were trigger orders, and so were 1 of 15 random unflagged fills and 2 of 18 unflagged fills picked as likely misses. So the trigger count is a lower bound as well. The file cannot tell a copy from a manual trade unless the app publishes a separate copy rate.
In the 30 days to 20 Nov 2026, most daily files ended between 12:00 and 12:20 UTC, some files of quiet builders held only fills after 12:25 UTC, and 4 Nov 2026 was the one whole day. Files 18 or more days old were still cut, so they are not completed later. A day counts as whole when its most active builder's file reaches 22:00 UTC.
The flag for trigger orders does not tell a stop-loss from a take-profit, and a stop sent without the app's builder code is not in its file. A zero means none was flagged among the fills carrying the code, not that the app has no stops.
Copies, small samples and missing rows
Two apps publish a separate rate for copies: HyperDash (5 bps on copy trades, 1.5 bps on its terminal) and Auspex (0.10% on the copies it places, 0.05% on orders you place). For them we also count the perp fills charged exactly that copy rate. That these are copies is an inference from the rate, not a copy flag.
A rate over fewer than 100 fills is not shown: the page prints a dash, and the count beside it stays.
A tool with no builder address in DefiLlama's registry has no on-chain row here. That does not show it sends no orders. DarkTrade's own row is not measurable this way until it publishes its builder address. Every figure is the app's own on-chain activity, not a forecast and not anyone's returns. We plan to repeat the measurement monthly.
Repeat it yourself
Each file is at stats-data.hyperliquid.xyz/Mainnet/builder_fills/<address>/<YYYYMMDD>.csv.lz4, one per builder per UTC day, with the address in lower case. An HTTP 403 means there is no file for that day. Each file is LZ4-compressed CSV with these columns: time, user, coin, side, px, sz, crossed, special_trade_type, tif, is_trigger, counterparty, closed_pnl, twap_id, builder_fee.
Markets follow DefiLlama's rule: a coin written #n is an outcome market, @n or X/Y is spot, dex:COIN is a HIP-3 perpetual, and anything else is Hyperliquid's main perpetual market.
The rows our pages show now cover 14–20 Nov 2026 and 22 Oct – 20 Nov 2026, measured .
Questions
Why do the comparison pages say 'at least'?
Hyperliquid's public builder-fill files cover only part of most days, so every count is of the published part. We never scale a count up to 24 hours. Rates, such as the fee and the share of trigger fills, are measured over the published part.
Does a zero trigger count mean an app has no stop-loss?
No. It means none of the fills carrying that app's builder code in the window was flagged as coming from a trigger order. A stop sent without the code is not in the file, the flag misses some triggered orders, and it does not tell a stop-loss from a take-profit.
Can you tell copied trades from manual ones?
Only where an app publishes a separate copy rate, as HyperDash and Auspex do: we then count the perp fills charged exactly that rate, which is an inference from the rate, not a copy flag. For every other app the file mixes both.
Why is DarkTrade's own row not measured?
DarkTrade is paid through a Hyperliquid builder fee, but it has not published its builder address yet, so its own row cannot be measured from the public files the way the others are.
How often is it measured?
We plan to repeat the measurement monthly. Every measured row on our pages prints its window and the day it was measured.
Sources
- stats-data.hyperliquid.xyz/Mainnet/builder_fills/<address>/<YYYYMMDD>.csv.lz4, read 21 Nov 2026
- github.com/DefiLlama/dimension-adapters/blob/f9003fdf9f77c483f7ca3cb1bd9815d1dcff572a/factory/hyperliquid.ts, read 21 Nov 2026
- hyperliquid.gitbook.io/hyperliquid-docs/trading/builder-codes, read 21 Nov 2026
- hyperdash.com/learn/hyperdash-vs-binance, read 21 Nov 2026 (HyperDash's copytrade rate, 5 bps (dated 17 Oct 2026))
- hyperdash.com/learn/hyperliquid-builder-codes-explained-how-third-party-apps-earn-fees-on-chain, read 21 Nov 2026 (HyperDash's terminal rate, 1.5 bps (dated 20 Jan 2026))
- auspex-trade.com/pricing, read 21 Nov 2026 (Auspex: 0.05% on orders you place, 0.10% on the copies it places)
Every figure is the app's own on-chain activity, not a forecast and not anyone's returns.