A daily loss limit is four decisions published as one number.

A daily loss limit is a level. It is calculated once, at the instant the trading day begins, and for the rest of that day your account is not permitted to be worth less than it. Everything else about the rule is composition:

limit = anchor − allowance
breached when the tested quantity falls below the limit

Four decisions are hiding in that sentence, and only one of them is on the marketing page. The anchor — which balance, taken at which instant. The allowance — a fixed number of points of the account you originally bought, or a share of whatever the anchor happens to be. The tested quantity — your realised balance, or your equity including everything currently open. The ending — whether touching the line closes the session or closes the account.

The published percentage sets the size of one of those four. It says nothing about the other three, and the other three decide whether any particular day survives.

This is measurable, and the measurement is unflattering. Of the 24 evaluation products in the rule corpus this site maintains, read on 2026-07-16, fifteen publish a daily loss limit and nine publish none at all. For every one of those fifteen, the corpus holds a dated quotation from the firm's own document for the headline percentage. For the ending it holds a quotation for seven. For what the limit is read on, three. For what the allowance is sized from, three. For the instant the day begins, none.

The number is published. The mechanic is not — and the mechanic is the part that decides your day.

The rest of this page is what those three unpublished decisions do, computed rather than asserted. Every level below is a percentage of the initial account, so the account you bought is 100.00% by construction and a limit of 5.00 points is a limit of 5%.

A daily limit needs a day, and a day needs an instant. Of the fifteen corpus rule sets that publish a limit, nine publish the instant it resets; six do not say. The nine that do name four different instants — the corpus records them in universal time and they are not the same moment.

Worse for anyone trying to be precise: a firm publishes a local clock time, and a local clock time is not a fixed instant. The corpus carries an explicit ambiguity note on exactly this — a reset published as an afternoon local time sits at one universal hour for part of the year and an hour earlier for the rest, because the firm publishes the local time and the clocks move underneath it. The boundary that decides which day your loss belongs to shifts twice a year, and nobody announces it.

Now the part that costs money. The anchor is a balance. The test is on equity. Those are two different quantities, read at the same instant, and one published worked example makes the arithmetic explicit: on the first day the anchor is the initial capital, 100.00%, and the limit is 95.00%; on the second day the anchor is the balance recorded at the rollover, 102.00%, and the limit is 97.00%. The allowance never moved. The anchor did, by exactly what the first day earned.

A position still open at the rollover is not in that balance — its profit or loss is not realised, so the anchor does not see it. It is in your equity, which is what the limit is tested against. So the new day opens below its own anchor, and the gap is charged to the new day in full, regardless of when it accrued.

On the session drawn above, the account rolls over at 108.00% with a position 0.60 points under water. The anchor is 108.00%; the limit, at a published 5% sized from the initial account, is 103.00%. The equity the session actually opens at is 107.40%. The allowance still standing is 4.40 points of a published 5.00.

An overnight position spends tomorrow's allowance before tomorrow starts, and it spends the whole floating loss, not the part that accrued after the boundary.

The reverse is equally true and rather less discussed: a position carried across the boundary in profit hands the new session more than its published allowance, because the equity it opens at sits above the anchor the limit was struck from. Neither direction is an exploit. Both are consequences of reading one rule off two quantities.

Both are published. In the corpus, eleven of the fifteen rule sets with a daily limit size the allowance from the initial account — a fixed number of points that never changes for the life of the account — and four size it from the balance the session opened at, so the allowance grows as the account grows and shrinks as it shrinks.

The two are the same rule on the first day of an account, exactly, because on the first day the opening balance is the initial account. They separate from the second day onward, and the size of the separation is not a matter of opinion:

separation = published percentage × (anchor − initial account)

Which is to say: the convention is worth nothing when you have made nothing, and worth most at the moment you have the most to lose. On the session drawn below the account is up, the two limits sit 0.40 points apart, and the session's low falls between them.

The low is 102.80%. Against the limit sized from the initial account, 103.00%, that is 0.20 points too far and the session is over. Against the limit sized from the opening balance, 102.60%, it is 0.20 points clear and nothing happens at all — the session goes on to close at 107.90%, 0.50 points above where its equity opened.

One session, one published percentage, one market. Under one convention it is a green day; under the other the account was flattened three quarters of the way through it.

Two of its readings are worth watching together. Session close is where the market took the session; into the next session is what the account actually kept. When they are the same number the rule cost nothing that day. When they are not, the distance between them is the price of the rule — not of the trade, and not of being wrong.

The instrument runs one shape of session — up, then down, then to the close — because that is the shape under which the conventions disagree most, and it says so rather than hiding it. Every reading it shows is recomputed in the page and checked on each build against model.js, an implementation that reaches the same answers by a different route: it never forms a limit level at all, and works instead in drawdowns from the anchor against the allowance.

Two of its four switches are conventions the corpus records. The third, the anchor, offers a level the corpus does not record any product using: a limit that hangs beneath the session's own high-water mark rather than beneath the rollover balance. It is there because traders routinely believe they are under one, and because seeing what it would do is the fastest way to understand what a daily limit is not. A limit anchored to the session high charges you for giving back gains made inside the session. A limit anchored to the rollover balance does not — it cannot even see them. And whichever it is, it forgets everything at the next boundary: the rule that remembers your high-water mark across days is the trailing floor, which is a different rule with a different memory.

Both are published, and it is close to an even split. Of the fifteen corpus rule sets with a daily limit, eight stop the session — no new orders until the next rollover, the account intact — and seven treat the same event as a failed evaluation. The corpus records an ending for fourteen of the fifteen and carries a direct quotation for seven of those; the remainder are inferences from rule text that does not say plainly which it means.

Under the ending that closes the session, the daily limit is not a killer at all. It is a throttle: it caps what a single session can cost and hands the account back. What it hands back is worth looking at. The session that stopped at 102.80% makes 102.80% the next session's anchor, so the next limit sits at 97.80% — the whole apparatus has moved down by 5.20 points and it does not move back up until the balance does.

That is the daily limit's entire memory: none of its own. It remembers nothing about yesterday except through the balance, and the balance is not a rule, it is your account. Which is what makes the two endings different products rather than different wordings of one. Under the throttle, the limit bounds a session and something else eventually ends the evaluation; the site's finding that the rule that binds need not be the floor is about which of them gets there first. Under the other ending, the daily limit is the terminal rule, and it has the shortest fuse in the contract.

The same published percentage, measured the same way, is a speed limit at eight of these fifteen rule sets and a cliff edge at seven of them.

• The counts are a reading, not an audit. They come from one corpus, read on 2026-07-16, of products a single reader could source. They are counts of products, not firms, and a firm's own current documentation is the authority over any of them.
• Four of the fields counted are transcriptions with no quotation behind them. Where the corpus records an ending, a measured quantity or an allowance basis that the firm's text does not state outright, it records an inference. This page reports both columns rather than the flattering one.
• Nothing here establishes how common each convention is in the market. A corpus of 24 products is a corpus of 24 products.
• The high-water-mark anchor is unobserved, not asserted. No product in this corpus is recorded as anchoring its daily limit to the session's own peak. The instrument offers it as a comparison so the difference can be seen, and it is labelled as such.
• Whether touching the limit exactly is a breach is not settled. Most rule texts say "below", some say "reaches", and the difference decides an exact tick. The instrument treats the limit as met when the tested quantity falls strictly below it.
• The instrument assumes a stop closes what is open, at the level it observed. Whether a rule set flattens positions or merely blocks new orders is stated by some texts and not by others, and a real platform's fill will not be the limit precisely. What the account carries forward is therefore the cleanest possible version of a stop, not the likeliest one.
• One shape of session is not every shape. Reverse the order — the low before the high — and the anchor conventions rank differently. The instrument holds the order fixed and names it.
• Levels here are percentages of the initial account. Real rule texts are written in currency, and their rounding, tick values and contract sizes are theirs, not ours.

Does being flat at the close protect me from the daily loss limit?

Not if the limit is read on equity, and every rule set in this corpus that publishes a daily limit is recorded as reading it on equity — that is, on your balance plus everything currently open. A position that goes far enough against you and then recovers has already been measured at its worst point. Closing the day flat protects your balance, not your day.

If I hold a position through the rollover, which day is the loss charged to?

The whole unrealised part standing at the rollover is charged to the day that follows it. The anchor for the new day is taken from your balance, which does not contain an open position's floating loss; the test is on your equity, which does. So the new day opens below its own anchor by the full amount of the float, and that much of the allowance is gone before the first trade.

Why do two accounts with the same published daily loss percentage disagree about the same session?

Because the percentage is a percentage of something, and the two rule sets disagree about what. Sized from the initial account it is a fixed number of points for the life of the account; sized from the balance the session opened at it grows and shrinks with the account. The two are identical on day one and separate by the published percentage multiplied by however far the account has moved since — which is why the convention matters most exactly when you have the most to lose.

Is a daily loss limit the same thing as a trailing drawdown?

No, and the difference is memory. A daily limit is recomputed at each rollover from a balance, and it forgets everything that happened before that instant. A trailing floor is driven by a high-water mark that ratchets and never moves down, so it remembers your best moment permanently. One of the corpus documents answers this question in as many words, for its own product, with a flat no.