Stop orders, and what they do not guarantee

A stop order does one thing reliably. Once the market has traded at or through the level written on it, the order is released and the position is closed. That is the whole of the guarantee, and it is a guarantee about an event rather than about a number. Everything else commonly attributed to a stop order is either a property of ordinary market conditions, which usually hold, or a promise nobody made.

The gap between what a stop guarantees and what it is assumed to guarantee is not a technicality. It is the specific place where a position sized on the assumption of a fixed maximum loss produces a loss larger than the one it was sized for, and it accounts for a large share of the outcomes that surprise account holders.

It does not guarantee a price 

The level on a stop order is a trigger condition, not a price boundary. When the condition is met the order converts into a market order, and a market order transacts at the best price available when it arrives. In a continuous market that price is normally at or very near the trigger, which is why the distinction is invisible most of the time. In a market that has moved between the trigger and the arrival, or that has jumped a distance without trading in between, the fill lands wherever the market is, and the difference is unbounded in principle.

The last row is not an extreme scenario invented for effect. It is the ordinary arithmetic of a market that reopens away from where it closed, and every stop resting inside that jump is filled on the far side of it for the same reason.

In one variant it does not guarantee a fill at all 

The variant is the stop limit, which converts into a limit order rather than a market order when it triggers. A limit order carries a price boundary, so where the market has already passed that boundary the order rests unfilled while the position stays open. The protection against a poor fill and the exposure to no fill are the same property, and no setting provides one without the other.

The confusion is compounded by naming. Some platforms describe a stop limit's two prices as a stop price and a limit price, others as a trigger and a worst acceptable price, and at least one convention places them in the reverse order in the ticket. Reading which field is which before submitting is the only reliable check, and the platform documentation states it.

It does not fire on what the chart shows 

A stop is evaluated against a live quote, and a market quotes two prices at once. A stop closing a long position is compared against the bid because closing a long means selling; a stop closing a short is compared against the ask. A chart drawn from one side of the quote will therefore appear to show a stop firing before the price reached it, or failing to fire after the price passed it, depending on which side the chart draws.

The width of the discrepancy is exactly the width of the spread at that moment, which means it is largest in precisely the conditions where a stop is most likely to be reached. This is a reporting mismatch rather than a fault, and it is inspectable: platforms allow the chart's price source to be switched between bid and ask so the two can be compared directly.

It does not cap what an account can lose 

A stop applies to one position. An account holding several correlated positions can have every stop triggered by the same move, and the sum of those losses is the account's exposure rather than any individual stop's level. Where the positions are correlated the events are not independent, so treating each stop as a separate bounded outcome understates the combined one.

Separately, an account can breach its margin requirement before any stop is reached, at which point the firm's close out procedure operates on its own terms and at the prevailing market price. That procedure is not the stop and does not consult it. A stop is one instruction among several that can close a position, and it has no priority over the others.

The instrument that does undertake a level 

A guaranteed stop is a separate instrument, not a setting. Where a firm offers one, it undertakes to close the position at the stated level regardless of where the market trades, and it absorbs the difference itself when the market opens beyond it. Because that undertaking is a real liability, it is priced accordingly: a premium is charged, availability is restricted to a published list of instruments, and a minimum distance from the current price is usually required.

Negative balance protection is a different mechanism addressing a different point in the sequence. Where a regulator requires it, an account's liability is limited to the funds in it after a catastrophic move, so a debit balance is written back. It operates after the positions have already closed at whatever prices the market offered, and it says nothing about those prices. Neither mechanism substitutes for the other, and an account can be covered by one and not the other.