Do Hyperliquid fee discounts survive third-party apps? Our mainnet test (October 2026)

1. The question, and why we asked it

Many traders never open app.hyperliquid.xyz. They trade through wallets and apps (Phantom, MetaMask, Based, Liquid, fomo and many others) that attach a builder code to each order and collect a builder fee on top of Hyperliquid’s own fee. A common claim, repeated on many referral pages, is that a builder code switches off the trader’s referral code for that order: the trader loses the 4% discount and the referrer earns nothing.

The claim has a real source. A Hyperliquid documentation page says “builder codes override referral codes for that order”. But that page describes a proposed staking referral program, and its header says there is “no definitive implementation” of it for mainnet. The live fee formula in the fees documentation applies (1 - activeReferralDiscount) with no builder condition. Documentation can lag behind the chain in either direction, so we measured what mainnet actually does.

We have a stake in the answer, and we were wrong about it once. Until October 1, 2026, some of our own pages repeated the “builder codes override referral codes” line. A first check on fills from September 29–30 contradicted it, and we corrected those pages. This study re-runs that check with a fresh, seeded sample, a fifth app, and a published script, so that anyone can verify the result without trusting us.

2. Data and method

All data is public and was read without an API key, with a generic User-Agent and no wallet connection.

Formulas. For each builder-coded fill: Hyperliquid’s part of the fee rate = (fee − builder fee) ÷ (price × size). Dividing it by the account’s own rate (taker if the fill crossed the spread, maker otherwise) gives a ratio: 1.00 means the full rate, 0.96 means a 4% discount. We take each account’s median ratio. For referrers: Δreward ÷ Σ(fee − builder fee) over the window, which would be 10% if the referrer earns 10% of Hyperliquid’s part, and Δreward ÷ Σ(total fee), which would be 10% if the builder fee counted too.

Reproduce it. Download the script (with its LZ4 decoder in a lib/ folder next to it) and run node hl-referral-builder-study.mjs --dates 2026-10-03,2026-10-04,2026-10-05 --sample 40 --delta-top 40 --wait-min 45 --seed cipher-2026-10 --out-dir out. It needs Node 22+, no dependencies and no key. The aggregate results are in this CSV (CC BY 4.0). Neither file contains a trader’s address.

Related:HypurrScan (hypurrscan.io): Hyperliquid Explorer for Any Wallet

3. The five apps in the sample

Builder addresses are public, but the apps do not publish which address is theirs. All five addresses carry an app label in DefiLlama’s open-source fee adapters, and we checked each with Hyperliquid’s API: three own a referral code named after the app (PURPS, SHIFU, INSILICO). fomo’s address owns a code with no referrals, and Liquid’s owns none, so Liquid rests on the label and its fee levels and is marked as inferred.

Plain USDC perps in the daily builder_fills files, October 3–5, 2026. How each address is attributed: Phantom: labelled “phantom-perps” by DefiLlama; the address itself owns the referral code PURPS; Based: labelled “based-app” by DefiLlama; the address itself owns the referral code SHIFU, which Based’s web app sets (read in its code on October 1, 2026); Liquid (inferred): inferred: labelled “liquid-perps” by DefiLlama, but the address owns no referral code; its 0.05% builder fee and 0.045% for referred users match Liquid’s published rates, and Liquid’s web app sets TRADEXYZ; Insilico: labelled “insilico” by DefiLlama; the address itself owns the referral code INSILICO; fomo: labelled “fomo-social-trading-perps” by DefiLlama; the address owns the code FOMOFAMILY (no referrals), and TRYFOMO, owned by another address, is the code every sampled fomo account carried, so we attribute it to fomo.

4. Which code do app accounts carry?

A Hyperliquid account has at most one referral code, set once. We read it for every sampled account:

The share of accounts carrying the app’s own code ranges from 100% (Phantom) to 2.5% (Insilico). That is consistent with apps setting their own code on accounts that arrive without one, which we also read in the code of the Based and Liquid web apps (October 1 check): Based sets its code only when the account has none, so accounts that arrive with a code keep it. Phantom is itself the wallet, and fomo creates the wallet, so their users rarely arrive with a code of their own. “Another code” covers every other referrer; we do not publish which ones.

5. Does the referee keep the discount on app orders?

Yes. Here is Hyperliquid’s part of the fee on builder-coded fills, divided by each account’s own rate:

Fills from October 4, 2026 to October 6, 2026. Accounts with no qualifying app fill in the window have no ratio. 12 accounts had a referrer but no active discount; 12 of them were above the $25M discount cap.

Every one of the 129 accounts with a discount had a median within ±0.5% of 0.96, and every one of the 21 accounts without one within ±0.5% of 1.00, whichever app sent the order. So for a tier-0 taker with a code, an app order costs 0.0432% for Hyperliquid plus the app’s builder fee, not 0.045% plus the builder fee.

6. Is the referrer paid on app orders?

Yes, on Hyperliquid’s part of the fee. Each dot pair below is one referee: the change in the referrer’s recorded reward for that referee, divided by Hyperliquid’s part of the fees the referee paid in the same window (blue) and by the total fee including the builder fee (orange). Rows are sorted by the builder fee’s share of the referee’s fees.

Window of 45 minutes on October 6, 2026. Not checked: 103 × referee not in the referrer’s list (the API returns at most 5,000 referees per referrer); 216 × under $1 of fees in the window; 20 × HIP-3 or spot fills in the window; 3 × referee above $1B lifetime volume (no reward by rule); 1 × traded only outside apps in the window (no builder-coded fill).

Every one of the 28 referees, all trading only through apps in the window, came out at 10.00% of Hyperliquid’s part, to two decimals, across all 5 apps. If the builder fee counted toward the reward, the orange dots would sit at 10%. They sit lower, and the further the more of a referee’s fees went to the app; the blue dots stay at 10% throughout. That matches Hyperliquid’s wording, “10% of referred users’ fees, less any fee discount they receive”, with “fees” meaning Hyperliquid’s fee, not the app’s. One side effect for anyone checking their own referrals: the API field cumRewardedFeesSinceReferred includes builder fees, so reward ÷ that field looks lower than 10% for app traders.

7. Limitations

• Sample. 200 accounts from 5 apps, one seed. Shares per app are estimates with wide error bars at 40 accounts each; the fee results (0.96 and 10%) are mechanical and showed no app-specific exception.
• Attribution. No app confirmed its address to us; the labels are DefiLlama’s, checked against the referral codes the addresses own. The builder we call Liquid owns no code and is inferred from the label and its fee levels. TRYFOMO is attributed to fomo because every sampled fomo account carries it, not because fomo documents it. MetaMask, Trust Wallet and other large builders are not in this sample.
• Timing. userFees returns the rate at the time of reading; a tier or staking change inside the window shows up as an outlier. Referrer snapshots are not atomic, so fills landing mid-request can shift a single referee’s ratio slightly; the totals are less affected.
• What we cannot see. The data shows which code an account carries, not who set it or when. We do not know how apps decide, beyond the web-app code we read for Based and Liquid. Hyperliquid can change any of this without notice; the dates above are when it was true.
• Privacy. We read individual accounts to compute aggregates, but publish no address, no individual referrer and no per-trader amount.

8. What it means for traders

1. Set your code first. Open a join link on app.hyperliquid.xyz with the wallet you will use, enable trading and deposit, before you connect that wallet to an app. Once an app has set its code, yours can no longer be added, and after $10,000 of volume no code can be entered at all.
2. Connect the same wallet. In apps that let you connect an external wallet (Based, Liquid), the account and its code come with you. An email or social login creates a new account without your code.
3. On fomo, use fomo’s code. fomo makes the account, so a Hyperliquid code from elsewhere cannot apply. Its own referral program cuts its fee.
4. Check your own account. The builder fee checker shows the code on any address and which apps can charge it. In any fill, (fee − builderFee) ÷ notional should be 0.96 × your rate if a discount is active.

Hyperliquid Builder Fee Checker — Paste a wallet address to see its referral code and every app approved to add a builder fee. Read-only, no wallet connection. — Check an address →

9. Disclosure and corrections

We run the Hyperliquid referral code HOLYGRAIL, so we benefit when readers use it, and a finding that codes survive app orders is good for us. That is why the method is fixed in advance by a seed, the script is downloadable and the data is public: the conclusions hold for any referral code, not just ours, and you can check them yourself. The study was commissioned, reviewed and published by BLTZ, the editor of this site, who is responsible for it. The data was collected by the published script; the analysis and the draft were prepared with automated (AI-assisted) tooling, and every number on this page is generated from the data file. If you find an error, write to us on X (@bltzman); corrections will be dated on this page.

Related corrections: on October 1, 2026 we removed the claim that app orders lose the referral discount from our referral code guide, the builder codes guide and other pages, after the first mainnet check.

FAQ

Disclosure: HOLYGRAIL is our Hyperliquid referral code; we earn 10% of Hyperliquid’s part of referred accounts’ fees at no cost to them. Data: Hyperliquid public endpoints, read October 6, 2026. Not investment advice.