How we verify

The short version

• Every fee, discount, limit and threshold our pages show comes from one facts registry: 4,833 records, each with the URL of the source we read it from (the exchange's or a government's own page wherever one exists) and the date we last checked it.
• Market figures such as volume and open interest are read from public APIs when the site is built, and shown with their date.
• Where documentation is unclear, we measure on mainnet with public data and publish the aggregate result.
• Our calculators are tested against each other on every automated deploy.
• Material mistakes are fixed on the page and listed in a public corrections log (17 entries so far, the latest on October 6, 2026).

One registry for every number

Numbers on this site are data, not prose. A fee such as Hyperliquid's base taker rate (0.045%) is stored once, with the page we read it from and the day we read it, and every article, table and calculator renders it from there. When an exchange changes a rate, we change one record and every page that cites it follows on the next build, so two pages cannot quietly disagree.

Records past their review date are listed on every build. An overdue critical record raises a warning on every automated deploy and fails our local pre-publish check until it is re-read at the source. Each check is logged with what was read, where and when, and pages show the date their figures were checked.

Which sources count

1. The exchange's own documentation, terms and public API. We read documentation pages in their raw form and query public endpoints directly.
2. The exchange's app, when the docs are silent or out of date (for example a withdrawal dialog or a referral rule that only the app states). We say so on the page.
3. Official government and regulator pages for tax and legal statements (for example Japan's Ministry of Finance and National Tax Agency, or Korea's FIU).
4. Secondary sources (news, law-firm notes, analytics sites) only when labelled as such, and never as the only basis for a fee or a legal claim.

If we cannot confirm something from a primary source, we write that it is unconfirmed instead of estimating it.

Live data, with dates

6 market figures, such as Hyperliquid's number of perp markets, 24-hour volume, open interest and the HLP vault, are fetched from public APIs each time the site is built (at least every six hours). If an API fails, the page shows the last saved value with its date rather than a guess, and the build records which sources were live in a public status file: /data-status.json. The build also re-reads Hyperliquid's USDC withdrawal route, which switched to CCTP in late September or early October 2026, and warns us if it changes again.

Our own measurements

Some answers are not in any document, so we measure them with public data:

• Referral discount on app orders (October 1, 2026): two days of public Hyperliquid fill data for four app builders, 120 accounts. It showed that the referral discount still applies to orders placed through apps, which contradicted what many guides (including ours) said. Corrections log entry · builder codes guide.
• USDC withdrawal route (October 6, 2026): 180 reads of Hyperliquid's routing endpoint plus the app's withdraw dialog. Withdrawal guide.
• Trading hours and weekends: statistics computed from Hyperliquid's hourly and daily BTC candles. Best time to trade · weekend trading.

We publish only aggregate results. We never publish individual users' wallet addresses or anything that profiles a person; addresses we name belong to apps or exchanges.

Screenshots we take ourselves

The 375 interface screenshots on this site were taken by us in a real browser, not taken from the exchanges' marketing material, and most carry a caption with the date we took them, so you can see what the screen looked like when we checked it. When an interface changes later, we keep the dated caption and note the change in the text.

Calculators tested against each other

Our trading calculators share their formulas and registry inputs, and 23 test files run before every automated deploy. Cross-tool tests check that the same input gives the same answer everywhere: the Hyperliquid referral discount across the fee calculators, liquidation prices across the liquidation, leverage and average-down tools, funding across the funding and PnL tools. A failing test stops that deploy.

When we get it wrong

We fix the page itself, change its "Updated" date when the fix is substantive, and add an entry to the corrections log with what the page said, what is right, the source and every page we fixed. When an exchange changes something, the log says how long our pages lagged.

Report an error: DM @bltzman on X with the page URL and what is wrong. A link to the source helps us fix it faster.

Referral links and editorial independence

This site earns money when readers join some exchanges and apps through our referral links; every code we use is listed on the About page. That is the only commercial relationship we have with them. Rankings and reviews are not sold, and holding a code does not change what we write about a venue's fees, limits or risks. We review venues that pay us nothing (dYdX, Antarctic, Pacifica), and we say plainly when our code does nothing for you: on Lighter Standard accounts, which pay 0% on regular orders, and on fomo, where a Hyperliquid code cannot apply.

Who is responsible

The site is run and edited by BLTZ (@bltzman), who is responsible for every page that carries the byline. More about the site: About Cipher Intelligence.