Precision and rounding

Every number in the settlement path is an integer. There is no floating point anywhere a value can affect a balance — because two machines computing the same floating-point expression can disagree in the last bit, and on a consensus execution path a last-bit disagreement is a fork. Prices and quantities are therefore held in fixed integer units, and every conversion between what you type and what the protocol stores has a defined rounding direction.

​

Tick size and lot size

Both are per market and published in the contract specification. An order that does not conform is rejected rather than silently adjusted — a venue that quietly rounded your price would be changing your order without telling you.

​

Rounding is a risk decision

Rounding direction is not a formatting preference. Each rounding choice allocates a fraction of error to someone, and the rule is consistent: where an error would create risk, round conservatively; where an error would be a systematic transfer, round fairly.

​

Rounded conservatively against the account

The pattern: anything that answers how much can this account take on rounds against the account. A margin requirement rounded down by a fraction, repeated across every position on the venue, is a systematic understatement of risk — small per instance, and exactly the kind of error that only surfaces when it matters.

​

Rounded fairly

These are the fields where a consistent direction would be a transfer rather than a safety margin. A fee that always rounded up would be a hidden surcharge; one that always rounded down would be a subsidy. Half-up leaves neither side systematically ahead. One exception: a fee that would round to zero is rounded up instead. A trade that pays no fee at all is worse than a trade that pays the smallest representable amount.

​

What this means in practice

• Conform to tick and lot size before submitting. Rejection for a non-conforming price is the most common avoidable order rejection.
• Do not size to the last unit. Free collateral rounds down and margin rounds up, so an order sized to exactly your available balance can be rejected. Leave a margin of error.
• Expect small display differences. Reproducing a value exactly requires applying the same rounding rule to the same integer inputs, not recomputing in floating point.
• Reconstruct from integers. If you are reconciling programmatically, work in the protocol’s units. Converting to a decimal type and back reintroduces exactly the ambiguity integers exist to remove.

​

Where to go next