price grid levels band edges band stop trailing stop wipeout distance (leverage only) band where it started
Only the as traded liquidation reading uses this. Exchanges set it per symbol and per position size — 0.4–0.5% is typical for large-cap perpetuals, higher for small ones. Moving it barely changes the answer; the number of levels held and the realised profit do.
What is being modelled. A long grid that starts flat: each level buys the first time price reaches it, and sells one level up. There is no initial position, so the opening days look quieter here than on a real bot that buys in at the start. Everything is counted in levels and percentages — no position size is entered anywhere, so no figure on this page is an amount of money.
Round trips and "still holding" are different things. Buying on the way down is not a completed trade. A configuration can fill hundreds of times and finish holding most of its levels, and the round-trip number on its own will not tell you that.
Direction decides which side the position builds on. Long buys each level on the way down and sells one level up. Short is the mirror — it sells each level on the way up and buys back one level down. Range runs both books at once over the same levels, so a move either way can complete a pair. The trailing stop is switched off in Range, because a peak that ratchets one way cannot follow two positions pointed in opposite directions. Band trailing is not — it works in all three.
The band stop sits outside the band, and it ends the run. Set it and everything closes the first time a bar reaches that far past the edge — below the low for a long book, above the high for a short one. It does not re-open: every bar after that is untraded, which is why the round-trip count can fall sharply when you switch it on. It can only fire if the band is narrower than the period, and the default band is drawn from the period, so narrow the range first.
The trailing stop follows the highest price seen since the run began. It tracks the highest high from the first bar (the lowest low, on a short) whether or not the grid is holding anything, so a grid that sells out at the top does not forget the top. It closes everything when price gives that percentage back — but only while a level is actually open, because closing nothing is not a trigger. It is not fixed to the band — it moves, and it only exists while the grid is holding something. The slider runs in half-percent steps to 50%. By default the run ends at the first trigger. Tick restart to let the grid re-open afterwards; the trigger count then tells you how often it would have fired.
Band trailing moves the band. It does not move what you are already holding. Switch on Trailing Up and the whole band steps up the first time a bar reaches above its top — the bottom rises with it. Trailing Down is the mirror. This is a different mechanism from the trailing stop above, which shares the word and nothing else: the stop closes positions, this one relocates the levels and closes nothing. A level bought before a move keeps the price it was bought at and the sell target it was given; the band can walk away and leave it there. That is what levels held outside the final band counts, and it is the number worth watching — a band that chases price down can look busy while quietly stranding everything it bought on the way.
Two rules, and neither is the right one. One level shifts the band by a single grid step per trigger. To the price it touched shifts it far enough to put that bar's extreme back on the edge, which on a daily candle can be a wick that lasted minutes. Run both. The difference between them is the size of the intrabar guess you are making. Trigger adds a buffer: at 0 the band moves the moment a bar reaches the edge, and at anything higher price has to clear the edge by that much first.
When one bar reaches past both edges, this does not move the band. A candle records the high and the low but not their order, so there is no way to say which edge was touched first, and the two answers move the band in opposite directions. Rather than pick one, this leaves the band alone and counts the bar under bars this data can't decide. A live bot would have moved on those bars. If that count is not zero, the run is not measuring your configuration — it is measuring the part of your configuration this timeframe can still see. This count only means anything with both directions switched on. With one direction on, a bar cannot reach past both edges in a way this has to decide between, so the count is zero by construction and says nothing about the data.
One more thing moves with the band. The band stop is defined as a distance outside the band, not as a fixed price, so when the band moves the stop moves too. A band trailing down carries its own stop down and can fall a long way without ever triggering it. If your exchange leaves the stop where you put it, this run is the more forgiving of the two.
Fee drag is a floor, not the total cost. It only counts the fees on trades that actually happened. Funding on perpetuals, financing or borrow on margined stocks and CFDs, roll cost on futures, swap on FX — all of those accrue while a position is held, which is exactly the stretch where a grid has stopped filling and fee drag can't see it.
The default band is drawn from the period you selected, which means price cannot leave it and the stop can never fire. That is the friendliest a configuration will ever look. Narrow the range by hand — that is where the interesting part starts, because a range chosen for the future will not contain the future.
Daily candles are a ceiling on what can be measured. If one candle's range spans several levels, this counts them all as filled, but the real path inside that day might not have hit them in that order. The finer the grid, the more inflated the count — that is what the bars-per-level warning is for. For fine grids you need shorter bars.
The wipeout line is optimistic, and it is drawn from the bottom of the band. At L× leverage a position is zeroed by roughly a 100/L% move against its average entry. This draws that distance down from the bottom of the band, which is the friendliest place to draw it from: a grid that filled higher up has a higher average entry, so its real liquidation sits at a higher price — nearer the band than the line shown, not further below it. Maintenance margin and accrued costs move it nearer still. Read the line as a floor on how much room there is, not a measurement of it.
Liquidation is asked twice, and the two answers are not interchangeable. Worst case assumes the grid is fully filled and no profit was ever kept: at L× that is a 100/L% adverse move, full stop. It is a screen for configurations that cannot survive, and it does not care what actually happened. As traded asks the same question of the position the run really carried — the levels that were actually open at each bar's extreme, against a balance that includes the profit booked up to that point. Both are computed in multiples of what you put in; no amount of money is entered anywhere, so none can come out.
The interesting case is worst case yes, as traded no. That is not “safe”. It means this configuration survived because it never filled all the way, and because every profit it made stayed in the account. Take the profits out along the way and the second reading moves toward the first. Read it as a conditional survival, and read the condition.
Where the as-traded reading is soft. It counts the profit this run booked, and those profits rest on fill counts this timeframe can only bound from above. More fills than really happened means a bigger balance than really existed. Pulling the other way, the fill model opens a level the first time a bar reaches it, including on the way up, while a real long grid rests its buy orders below the price and would carry fewer levels — which makes the reading stricter. The two errors point in opposite directions and this tool cannot say which one wins on your data. That is why the worst-case line stays.
The as-traded reading is answered at 1× too. Worst case has nothing to say without leverage — 100/1 is a 100% move — but an account can still be emptied without borrowing: a short loses more than everything it put up if price more than doubles. A long at 1× cannot, and the reading will say so.
In Range the as-traded reading is not shown. An exchange nets a buy book and a sell book against one pool of margin. This tool keeps them as two separate books, which is right for counting fills and wrong for counting margin, so it declines to answer rather than answer wrongly.
Two things here are called liquidation, and they are not measured the same way. The red line on the chart is the one just described — drawn down from the bottom of the band, optimistic by construction. The liquidation figure in the metrics is not that line. It compares the deepest a level went underwater from its own entry against the 100/L% distance, which is what the leverage rule actually asks for. A level can pass that distance while price never leaves the band, and when it does, this reads yes. Both are approximations — the line is optimistic, the depth is an upper bound — so read a yes as “this configuration was not survivable on this window”, not as an exact liquidation price.
"Worst move below the band" is the one number here that does not depend on intrabar path. It only needs the extreme low of each bar, so the timeframe warning does not apply to it. That is why it is worth more than the fill counts.
"Deepest underwater" is an upper bound, not a measurement. On the bar where a level is bought, there is no way to know from a candle whether that bar's low came before the fill or after it. This counts it as after — the pessimistic reading — so on that first bar the figure can overstate. From the next bar on it is exact.
My own grid: 5 grids, 90 days, 2 trades closed →