What Reduce-Only Does Not Cover
Two fields on the order form look like they do the same job. reduce_only promises your order cannot grow the position; close is the one that actually ends it. Confusing them costs a little. What costs more is assuming the reduce-only stop you left on a position weeks ago is still standing after you trimmed that position by hand. This page takes the field distinction first because it is the one you can settle in ten seconds, then the assumption, which is the one that empties accounts.
Two fields, two jobs
Gate's order endpoint carries reduce_only and close side by side, and the documentation gives each a different job.
For close it says to set the flag true and the size to 0. The size field repeats the instruction from its own side: size is required, and 0 is the value a close-position order carries. Nobody works out a quantity; the venue reads it off the position.
Reduce-only never takes that job. You still name the quantity. All the flag adds is a promise about what that quantity cannot do, which is grow the position.
Follow that through and an asymmetry falls out. Type a size larger than the position and the flag has something to say. Type one smaller and it has nothing to say at all, because a smaller number was permitted from the start.
| If you type | Reduce-only | Close |
|---|---|---|
| More than the position | The excess is not allowed to open anything | Not applicable; you type 0 |
| Less than the position | Accepted in full, silently | Not applicable; you type 0 |
| Nothing at all | Not a valid order | This is the normal case |
Undershooting is the error worth guarding against, because it is the one people actually make. Partial exits invite it: sizing off a currency figure, or forgetting an earlier trim. Both leave you believing a position is nearly gone when most of it is still open, and nothing on screen contradicts you.
The narrow thing it guarantees
Stated plainly, the flag covers one instant: the moment the order is accepted. After that its guarantee is spent.
The documentation is correspondingly brief. The field means the order is reduce-only; the notes above the parameter table add that only position-reducing operations should go through, so the position cannot increase. Every exchange help centre explains that much, which is why this page does not.
The assumption that costs
Here is the failure that actually empties accounts, and it has nothing to do with typing.
A stop goes on a position in week one. In week two the position is trimmed by hand, or closed and reopened. In week three the market moves, and the stop is assumed to be watching.
It may not be. Sized against a position that no longer exists in that form, it stopped corresponding to reality the moment you touched the position, and anything reopened afterwards is a new position as far as the venue is concerned.
What the documentation offers is two order-completion values. One reads: cancelled because the position was closed. The other: cancelled because of increasing position while reduce-only is set. Between them they establish that an order like this can end in cancellation rather than patient waiting.
What they do not establish is the reverse — nowhere is there a promise that closing a position sweeps every related order away for you. Which leaves both assumptions unsafe. Counting on the order still being there may be wrong. Counting on the venue to have tidied up may also be wrong.
My own rule follows from that rather than from the docs: I do not trust the sentence "I remember placing one". Memory is suspiciously cheap here, and the order list costs nothing to open.
None of this makes a sound, which is the part that gets people. A line disappears from the order list, or a status word changes. Our page on a stop that did not trigger or filled far off files this under orders that stopped mattering; it sorts faults after the event, where this page is about not reaching one.
Hedge mode adds a side
Under hedge mode the problem above acquires an extra dimension.
One-way mode holds a single direction per contract, so reducing is unambiguous. Hedge mode lets long and short coexist, and reducing therefore has to carry a direction: reduce_only true with the sign of the size naming the side, or, to close one side outright, size 0 plus a field that names close_long or close_short, with reduce_only still set.
None of which you will type by hand. The reason it matters is that a resting reduce-only order can end up attached to the side you are not worried about — most easily when both sides are open, or just after a mode switch. Nothing fills, and nothing errors.
Acting from the positions row sidesteps it. A control attached to a position already knows which position it means, whereas the order form asks you to get direction, size and the flag right simultaneously.
One habit
Mechanism aside, one action changes outcomes, and it is close to free: after touching a position by hand, open the order list.
Two questions there. How much of the position is actually left, and does the resting order carry that number? In hedge mode, add a glance at which side it is sitting on. Answers that line up mean the position genuinely has cover.
The boundary, so that nothing is expected of the flag that it never offered:
| Concern | Covered? |
|---|---|
| Oversized exit turning into a reverse position | Yes |
| Undersized exit you believed was the whole thing | No |
| Position changed after the order was placed | No; the order may already be cancelled |
| Resting on the wrong side under hedge mode | No, and nothing errors |
| Whether it fills, and at what price | No; that is depth and order type |
Field names, where the option sits and the exact validation vary by venue and by version, so the order form in front of you and your own order history are what count. Only Gate's public documentation was checked here; how other venues handle this step was not looked at and is not inferred. The mechanics come from Gate's API v4 futures documentation (the order endpoint and the order finish statuses), with the help centre used only to confirm what the order-form options are called. Checked 2026-09-07; no account was signed into, no order was placed, and no account screen was captured while writing it.
Whatever you decide about the checkbox, position size and stop placement remain the two things that set what a trade costs you: risk management and setting stops.