Privacy

You are handing a trading analytics tool your complete trade history. It would be reasonable to want to know exactly what happens to it. This page is specific rather than reassuring, because a vague promise is worth nothing here.

There is no account

No sign-up, no email, no password, no wallet connection, no signature request. You never authenticate, so there is no identity for anything to be attached to. Reading a public address is read-only by construction — nothing here can move funds or place an order, and we never ask you to approve anything.

Never stored, at all

Your uploaded file. Parsed in memory, never written to disk.
Your wallet address. Used for the venue request, then discarded.
Any individual trade, fill, price, size, symbol or timestamp.
Your Lighter auth token. Used for that one request, never logged.
Cookies, analytics, fingerprints, or an account — there isn't a login.

Your report is generated in a single request and lives only in your browser tab. Close it and it is gone — there is nothing on our side to delete, and no way for us to retrieve it.

The one thing that is kept

We would rather name it than claim a clean “we store nothing” that isn't true. Percentile rankings need a population to rank against, so each report appends four rounded numbers to an anonymous pool:

Cost per $1,000 of notional traded
Maker/taker ratio
Loss/win size ratio
Profit factor

Four rounded numbers per report, with no address, no timestamp and no link between them and you. They exist only so the next person can be told where they rank. You can switch this off and still see your own percentiles.

They are written as four bare numbers with no address, no timestamp, no venue and no identifier — nothing that links a row to a person, and nothing from which any trade could be reconstructed. Accounts with fewer than ten closed trades are skipped entirely. If you would rather not contribute, there is a toggle on the upload form; your own percentiles still work.

Who else sees your data

Hyperliquid and Lighter. When you paste an address we request its history from their public APIs. They see the request, as they would for anyone reading public chain data.

OpenAI. Only for layouts we can't parse ourselves — an unfamiliar CSV header, a PDF statement, a screenshot. In that case the file contents are sent to their API to identify columns or transcribe a table. Recognised exports from Hyperliquid, Lighter, Binance, Bybit, OKX and dYdX never touch it, and neither does an address lookup. The model is never asked to compute a number, only to read a layout.

Nobody else. No analytics, no advertising, no third-party scripts, no tracking pixels, no data sales.

Verify it yourself

Claims like these are easy to make. The relevant code is small and readable: src/lib/cohort.ts is the only file that writes anything, and it takes exactly the four values above. Open your browser's network tab while you run a report and you will see one request out and one response back.

Questions or something that looks wrong — say so and it gets fixed. If this policy ever changes materially, it will change here first, and the change will be described rather than quietly reworded.