Hyperliquid frontends builder revenue leaderboard

The bench ranks every Hyperliquid frontend with a known builder address by the USD value of builder fees they collected on the last complete UTC day, and over the 7 and 30 days ending on it. Source data is the per-builder daily fills feed Hyperliquid publishes at stats-data.hyperliquid.xyz: one lz4-compressed CSV per builder address and UTC day, listing every fill that carried that builder code with its notional, wallet, taker flag, closed PnL and builder fee. A Go harness on OCB infrastructure downloads each file once the day is published, keeps a 30 day local mirror, sums fees and notional per builder for the day, 7 day and 30 day windows, unions the daily wallet sets for the user counts, and exposes the totals as Prometheus gauges (hl_frontend_fees_usd_24h_v2, _7d_v2, _30d_v2). This page reads the gauges every 30 seconds. The registry of tracked builder addresses lives at `harnesses/hyperliquid-frontends/builders.json` in this repo, seeded from DefiLlama dimension adapters and HyperTracker plus a cross reference pass on the top unidentified addresses. New builders ship via public PR. The bench does not place trades and does not touch private keys. Two sources supply the days. Hyperliquid's public export has been cutting most days off around 12:10 UTC since 2026-09-08, and on a day it did carry in full the hours before 13:00 hold roughly 43 per cent of the notional, so a window summed from it alone lands near half of itself. Our own Hyperliquid node writes every hour of a day it was up, and where it holds a day end to end that day replaces the export's copy outright. The two are never averaged: two readings of one quantity that differ by a factor of two are not two readings. A day the node cannot replay may instead be reconstructed from the export's own morning spliced to an afternoon rebuilt wallet by wallet from the info API, and a reconstruction is counted and reported apart from a measurement. The coverage note on the Hyperliquid hub states how the current window divides between the three.

• ·Data source. Hyperliquid publishes one CSV per builder address and UTC day at stats-data.hyperliquid.xyz/Mainnet/builder_fills/<address>/<YYYYMMDD>.csv.lz4, listing every fill that carried that builder code with time, wallet, coin, price, size, taker flag, closed PnL and builder fee in USDC. The feed is public and keyless; a missing file answers 403.
• ·Cadence. Day D is published early on D+1 (observed 02:45 to 03:30 UTC). The harness re-checks the feed every 30 minutes, downloads new files into a local mirror, and switches the whole cohort to day D once at least five builders have a file and the batch has settled for 45 minutes. Prometheus scrapes the gauges every 30 seconds. End-to-end lag from a fill to this page is 3 to 27 hours.
• ·Headline calculation. For each builder, the sum of builder_fee over every fill in the day-D file. The 7 day and 30 day figures sum the same field over the files of the 7 and 30 UTC days ending on D, so the three ranges are consistent and every builder is compared on the same days.
• ·Volume companion. Notional is px times sz per fill, summed the same way for the day, 7 day and 30 day windows.
• ·Users. Distinct wallets in the day-D file. The 7 day and 30 day counts are unions of the daily wallet sets, never sums, so a wallet active every day counts once.
• ·Effective fee bps. Builder fees divided by notional, times ten thousand, on day D. The trader perspective on what a representative dollar of flow paid the frontend.
• ·Builder registry. builders.json in the harness tree is a hand-curated array of slug, name and address for every tracked frontend. A frontend routing through several builder addresses can list them all under one entry and they are merged into one row; MetaMask's documented secondary address stays its own row (MetaMask alt) for continuity with the recorded history. New addresses ship via public PR. Addresses outside the registry are not fetched.
• ·Exclusions. Native Hyperliquid UI orders carry no builder code and are not attributable; roughly 95 percent of Hyperliquid fills land without one. HIP-3 market fills routed through a frontend count toward that frontend; deployer fees are not part of this feed.
• ·Node substitution. Hyperliquid's public export cuts most days off around 12:10 UTC, so where our own Hyperliquid node holds a day end to end that day replaces the export's copy of it outright, never an average of the two. A day the node cannot replay may be reconstructed from the export's morning plus an afternoon rebuilt per wallet from the info API; reconstructions are counted and published apart from measured days, because close is not measured.
• ·Reproducibility. Harness source is open in the OCB harness tree. Anyone can run it against the same public feed with the same registry, which reproduces every number for the days that rest on the export. The days our node measured cannot be reproduced from the public feed alone, and the coverage gauges say which days those are.
• ·Failures. A transport error leaves the previous file and gauge values in place and is retried on the next pass. The cohort never switches to a day before its batch has settled, and the choice of day only moves forward. Restarts re-read the local mirror before touching the feed.