How We Test & Rate Perp Exchanges

TL;DR — the rules we hold ourselves to

• Every number on this site comes from a verifiable public source (the exchange's own API or an independent aggregator), never from self-reported marketing material.
• Fees, spreads, and slippage are recomputed from live order books every 60 seconds.
• Audits are linked to the actual report PDF on the auditor's domain, not to the exchange's claim that one exists.
• Affiliate partnerships do not raise ratings. Affiliate-link absence does not lower them.
• If we cannot honestly verify a number, we either omit the venue or label the data as "observed" with a date — we never fabricate.

Our rating dimensions

Every perp exchange — DEX and CEX — is scored across five dimensions on a 0–10 scale. The composite "Overall" rating is a weighted average. Sub-scores are visible on each review page so readers can re-weight by what matters to them.

How we measure spread and slippage

For each supported venue we fetch the live BTC/ETH/SOL/XRP perp order book directly from the venue's public API, request the server's maximum allowed depth, and compute:

• Spread = (best ask − best bid) ÷ mid × 10,000 bps
• Slippage = (VWAP at trade size − best ask) ÷ best ask × 10,000 bps
• Total cost = fee + half-spread + slippage

We exclude order-book levels priced more than 5% from the best ask. Some venues' public feeds include stale or spam limit orders far from mid; a real market order would never fill there. Including them inflated reported slippage by hundreds of bps on a small set of venues. The deviation guard makes large-size slippage reflect realistic executable liquidity.

For oracle-priced venues (Jupiter, GMX, Gains, Reya, Avantis, Ostium, Kwenta, MYX, GMTrade) the order-book model does not apply — trades fill at the oracle reference price with no spread or book-depth slippage. These appear with spread = 0 and slippage = 0, and their cost equals the fee alone. We never apply this treatment unless the venue's execution model is genuinely oracle-based.

What we don't (yet) verify

We are explicit about gaps so readers can calibrate trust:

• RFQ-spread venues (e.g. Variational) where the spread is per-quote and not in a public book — we exclude these from the cost-comparison tool rather than show a misleading 0-cost number.
• Venues that require wallet-signature auth for order-book data and have no public broker mirror.
• Order-book contract multipliers that the venue does not document publicly — we hold the venue out of the comparison until the multiplier can be verified.
• Self-reported reserve attestations that are not Merkle-tree verifiable per-user.

Data sources we use

Editorial independence

PerpFinder carries affiliate links to some venues. We disclose this on every page that contains one. Our rating methodology is identical for affiliated and unaffiliated venues and is applied before any commercial relationship is considered.

If a venue we have an affiliate deal with rates poorly on our methodology, we say so. The cost-comparison tool ranks by cost, not by referral payout. Where a venue cannot meet our data-integrity bar (e.g. missing audit, unverified contract multiplier, RFQ-only spread), we exclude it rather than fill the gap with estimates.

Corrections policy

Find a number that doesn't match the venue's public API or docs? Email contact@perpfinder.com with the URL and the discrepancy. We fix verifiable errors within 48 hours and update the About page change-log.