Our cheap orders were costing us money on every exit

Exchanges charge less when your order waits in the book (a "maker" order) than when it takes an existing price (a "taker" order). On Binance futures the difference is roughly 2 basis points, 0.02% of each trade. For a bot that turns its account over dozens of times a year, that adds up. So we taught the bot to post a maker order first, wait 45 seconds, then take the market for anything left over.

Two days on Binance Demo Trading showed it was right for entries and wrong for exits.

What the fills showed

We compared every demo fill with Binance's own record of whether it was maker or taker, and with the real market price at the same minute.

• The demo order book is realistic. Fills landed within ±5 bps of the real market price at the same minute.
• Entries worked. The one new position in that window filled 74% as maker, at +0.7 bps from the price the bot saw when it decided.
• Exits didn't. Every trim and close got 0% maker fills and ended 9 to 22 bps worse than the price at the moment of decision.

Why

A trend bot exits when the price is moving against the position. A passive order to sell sits just above a falling price and doesn't fill. After 45 seconds the bot gives up and sells at market, now lower than when it started waiting. You pay the full cost of waiting to save a fee you never get.

Entries are different: the bot buys into a move that is going its way, with no hurry, and a passive order at the bid often fills while the price wobbles.

What we changed

Entries and additions keep maker-first. Trims, closes and stop-losses now go straight to the market. Every fill is still logged with its maker share and its slippage, and the weekly report shows both, so we'll know if this needs changing again.

It's a small fix, but it's exactly why a strategy runs on a demo account before real money: no backtest would have told us this.