Leveraged Grid Bots: What the Profit Card Leaves Out

A leveraged grid bot card showing "+6% in 8 days" usually measures the profit on the margin committed, over a window where price stayed inside the grid. It does not show the real position size, how far away liquidation was, what funding cost, or how the same grid would have done in other weeks.

Published 27 October 2026. A generic, fully worked example; no real card is reproduced. Research on past data, not investment advice.

What a grid bot card shows

OKX, Binance and Bybit all offer grid bots on perpetual contracts, with a leverage of your choice. The share card of such a bot usually shows:

• a profit percentage, often relative to the margin committed (the "investment");
• a profit in currency (USDT or USDC);
• a running time;
• sometimes an annualized yield, the direction (long or short) and the leverage.

These figures are correct for the window shown. What is missing is everything you would need to judge the risk taken to get them.

The example: a generic grid, worked out end to end

The figures below are round so you can check them by hand; the Python calculation is given at the end of the article.

1. The percentage is on margin, not on the position

+60 USDT on 1,000 USDT of margin is +6%. Measured against the 8,000 USDT of notional the grid can carry, it is 0.75%. An 8x leverage multiplies the displayed percentage by 8, both ways.

For a grid, profit comes from round trips: one 1,000 step near 100,000 earns 10.10 USDT gross, minus 0.40 USDT of fees at 0.02% per side. It takes about 6.2 full round trips to make 60 USDT. The card rewards a window where price oscillated without leaving the grid.

2. Liquidation gets closer as the grid fills

A long grid buys the dip. If price falls to the 92,000 floor, all 8 buys are filled:

• position: 0.08382 BTC, average entry price 95,445;
• unrealized loss at the floor: −289 USDT, that is −28.9% of margin for a price 8% below the start;
• effective leverage at that point: 10.8, more than the leverage chosen, because margin has shrunk;
• liquidation price: 83,850, that is 8.9% below the grid floor and 16.2% below the start price.

For comparison, the same margin committed all at once at the start price with 8x leverage would be liquidated at 87,851, 12.1% lower. The grid moves liquidation further from the start price, but a 9% break below the grid is enough to lose everything.

Exchange liquidation figures differ slightly: maintenance margin tiers, liquidation fees, mark price instead of last price. The order of magnitude stays the same.

3. Funding is paid every 8 hours

A long perpetual pays funding when the rate is positive. At 0.01% per 8-hour period (a common base level on BTC perpetuals, not a forecast), 8 days make 24 periods:

• grid half filled on average (4,000 USDT of notional): 9.6 USDT;
• full grid (8,000 USDT): 19.2 USDT.

That is 16 to 32% of the 60 USDT gain. Check whether the card's figure includes funding and fees. Real rates change every 8 hours, rise markedly in bull markets and can turn negative.

4. Time spent inside the grid

A grid only earns while price stays between its bounds. Above, it has sold everything and does nothing; below, it carries its full leveraged position. An 8-day card shows a window where price stayed inside. The right question is: over the history, what share of the time would the same grid have spent inside its band, and how many launches would have ended liquidated?

5. No annualization under 90 days

+6% in 8 days, annualized, gives 273.8% simple (×365/8) and about 1,328% compounded. These figures assume the same window repeats 45 times in a row, with no grid exit, no liquidation and the same funding. They describe an assumption. The Lab brings a card's figure back to its own window and never annualizes a window shorter than 90 days.

Questions to ask before believing a card

Testing a grid on history with the Lab

Strategy Arena Lab 0.1.0-252, free for Windows and Linux, already replays a long grid with leverage from 1 to 125 on past data through its local MCP server (your AI sends the settings): maintenance margin, liquidation, liquidation fee, dated Binance or OKX funding every 8 hours, worst drawdown, comparison with holding, and the exchange's figure brought back to its window. The exchange grid judge, which reads the settings of a card and re-judges the same grid over the whole history, is coming in the next version of the Lab.

Limits of 0.1.0-252, written in every result: liquidation reads candle prices (stricter than the mark price), the embedded funding table ends on 26 October 2026 at 08:00 UTC (beyond it a leveraged run is refused, never invented), and the computation runs on the CPU. The Lab places no orders and connects to no exchange account.

FAQ

What does the percentage on a leveraged grid bot card measure?

Usually the profit relative to the margin committed, not to the position size. At 8x leverage, a 6% gain on margin is 0.75% of a full position.

Why not annualize a gain made in a few days?

Because annualizing assumes the same window repeats all year. Over 8 days, +6% becomes 273.8% simple and about 1,328% compounded: those figures describe an assumption, not a result. The Lab never annualizes a window shorter than 90 days.

Where is the liquidation price of a leveraged long grid?

It moves down as buy orders fill, but effective leverage goes up. In the article's example (8x leverage, grid from 100,000 down to 92,000), liquidation sits at 83,850 once the grid is full, 8.9% below the grid floor.

Does perpetual funding change the result?

Yes for a position held several days. At 0.01% per 8-hour period, 8 days cost between 9.6 and 19.2 USDT on 8,000 USDT of notional, that is 16 to 32% of a 60 USDT gain. Real rates vary and can be higher or negative.

Can the Lab test an exchange grid?

Lab 0.1.0-252, free for Windows and Linux, replays a long grid with leverage from 1 to 125 on past data through its local MCP server: maintenance margin, liquidation, liquidation fee and dated Binance or OKX funding. It places no orders.

Sources and calculations

• Example calculation, to reproduce in Python (standard library; liquidation fees are ignored):

margin, lev, mmr = 1000, 8, 0.004
levels = [100_000 - 1000 * (i + 1) for i in range(8)]    # 99,000 ... 92,000
notional = margin * lev                                    # 8,000 USDT
btc = sum(notional / 8 / p for p in levels)                # grid full
p_liq = (notional - margin) / (btc * (1 - mmr))            # equity = mmr x position
print(round(btc, 5), round(notional / btc), round(p_liq))           # 0.08382 95445 83850
print(round((92_000 - p_liq) / 92_000 * 100, 1))                    # 8.9
print(round(6 * 365 / 8, 1), round((1.06 ** (365 / 8) - 1) * 100))  # 273.8 1328
print(round(24 * 0.0001 * 4000, 1), round(24 * 0.0001 * 8000, 1))   # 9.6 19.2

• Liquidation of a leveraged long: equity brought down to maintenance margin, as in Lab 0.1.0-252 (published 27 October 2026). Lab defaults, read on 26 October 2026 from the exchanges' public parameters: maintenance margin 0.4% (OKX BTC-USDT tier 1) and liquidation fee 1.25% (Binance BTCUSDT, ignored in the example above).
• What changes in Lab 0.1.0-252.
• Further reading: why a Pine strategy wins in backtest and loses live, plug your AI into the Lab over MCP.