Last Updated: 2026-10-27
An advertised leverage ceiling does not tell you how a contract settles, which collateral is exposed or what happens when an exit fails. Neither does a large order book establish account eligibility or custody safety. Use the worksheet below to compare the same proposed exposure across venues. There is no measured venue ranking in this guide.
Start with hard constraints
Record your jurisdiction, the contracting entity, the exact product and the provider's current access terms. An account that can use spot may not be eligible for derivatives. If eligibility is unresolved, stop the comparison there; a cheaper fee cannot compensate for it. Do not use location masking or a different identity to evade restrictions.
Next describe the contract. Linear, inverse, perpetual and dated contracts are not interchangeable. Identify the quantity unit, settlement asset, expiry or funding schedule, collateral rules and position mode. Stablecoin settlement is not settlement in a risk-free bank dollar. A hedge in one contract can leave basis, currency or settlement risk against another.
Scroll sideways to see every column. Keyboard: focus the table region and use the arrow keys.
| Requirement | Evidence to save | Unresolved case |
|---|---|---|
| Eligibility | Current entity/product terms and date | Do not assume access from a signup page |
| Contract | Symbol specification, quantity unit, settlement and expiry | Do not compare unmatched exposures |
| Collateral | Margin mode, auto-add settings, eligible collateral and haircuts | Do not infer an isolated loss cap from a label |
| Execution | Timestamped depth for the same quantity, both directions | Insufficient depth is a result, not a missing row to omit |
| Orders | Trigger reference, reduce-only/close behavior, partial-fill handling | A stop displayed on screen is not proof of protection |
| Exceptional closure | Liquidation, insurance and auto-deleveraging rules | An insurance fund is not deposit insurance |
| Costs | Filled notional, fee tier, funding events, conversion/transfer costs | No overall winner until units and scope match |
Provider documentation is a description of intended behavior, not evidence that a particular account or stress event worked correctly. Keep observed logs separate from those descriptions.
Preserve the useful fee-versus-fill comparison
At $10,000 of filled notional, hypothetical fees of 0.055% and 0.060% differ by $0.50 for one leg. A 0.5% adverse execution difference on that same notional is $50. This shows why both fields matter; it does not establish that any named venue has better execution, or that fees are always trivial.
For a reproducible depth study, record both books at a stated UTC time, calculate the weighted fill for the same quantity, and retain insufficient-depth and stale-feed cases. Repeat across a predefined observation window. Do not label a normal-session snapshot a stress test. A replay that assumes all displayed orders remain available can understate adverse selection and cancellation risk.
An actual execution comparison would also need acknowledgement and fill timestamps, rejected/cancelled orders, account settings and the full selection method. Without those records, phrases such as "never disconnects" or "keeps matching during every crash" are unsupported.
Keep funding on its own ledger
Synthetic example: unchanged $10,000 notional, a payment of 0.05% at each eight-hour event, ninety events over thirty days. Total payments are $450, or 4.5% of notional. A one-time 0.055% entry fee is $5.50; the funding total is about 81.82 times that fee, not fifty times. This is constant-rate arithmetic, not a likely monthly expense.
For a real record, sum each signed payment using the actual position and rate at that event. Payments may be received rather than paid, and displayed estimates can differ from settled rates. Binance's market-data documentation separately exposes funding history and funding-interval information; do not assume every instrument permanently settles every eight hours.
Dated futures avoid a perpetual funding schedule but introduce expiry, basis and roll considerations. That distinction alone does not make either product suitable for a particular holding period.
Separate trigger, execution and forced closure
The last traded price, a calculated mark price and an executable bid or ask can differ. Identify which reference controls your conditional order and which controls liquidation. Selecting the same reference for both does not ensure the exit fills before forced closure. Binance's stop-order guide describes this distinction and the non-fill risk of stop-limit orders.
Do not use 1 / leverage as an exact liquidation distance. Maintenance tiers, fees, collateral, account mode, other positions and payments affect the calculation. Likewise, "isolated" is not a universal promise that no other funds can be added or exposed: verify the applicable settings and terms.
For each candidate, write down the documented auto-deleveraging process and what record would show it occurred. Closing a profitable position through such a process is different from a stop, a voluntary close or a platform outage. Compare the consequences rather than treating a larger published insurance balance as proof that the event cannot happen.
Finish with a bounded conclusion
An honest result might read: "For this instrument and quantity, these six dated snapshots showed lower modeled execution cost at A; eligibility and stressed-order handling remain unresolved." It should not become "A is the best exchange" without broader evidence.
Keep compulsory requirements separate from preferences. A missing required settlement asset or an unavailable product disqualifies a route; it does not merely subtract a point from a weighted score. Save a no-suitable-route outcome when appropriate.
For a worked margin and fill-state exercise, use the futures learning worksheet. For cost units, use the fee comparison worksheet.
Educational comparison method, not a recommendation to trade derivatives. Losses can consume collateral and may exceed the initial margin depending on the product and account. All numerical examples are invented. No funded platform tests or personal returns are claimed. The site may earn referral commissions elsewhere; no commission determines this worksheet's criteria.