The short answer
Pick the way a position is built and left by the market you expect, then measure it on the venue's history before any real order. Four ways cover most of what a program trades:
- By steps, for a market you expect to come back: buy a little, then more lower, and leave at the average price plus a target.
- With a stop, for a market that runs: one entry, a stop decided before it opens, and a target or a stop that follows.
- A grid, for a market that goes back and forth in a band: buy under the price, sell over it, and each round trip pays one step.
- A range, for a quiet market: two entries at the borders of the last hours, the first filled cancelling the other.
By steps
Steps posted in advance under the price, in the lower part of the week's range, each a little larger than the one before; the whole position leaves at its average price plus a target. There is no stop: what protects the account is the room left before liquidation, which sets the size of the base stake.
Two variants keep the idea: a DCA buys the same stake at the market every few days whatever the price, and an exit in parts takes a share at the average plus 0.5 %, then 1 %, then 1.5 %. Leaving in parts feels safer; measured over two years it returned less than leaving at once.
With a stop
One position at a time, its stop placed before it opens, and its size chosen so that the stop loses a fixed share of the account. The exit is a target, a stop that follows the price and never comes back, or a part taken first and the rest followed.
What decides the entry changes the strategy: a breakout past the high or the low of the last thirty days, a pullback in a strong trend, an excess of the hourly RSI in a range, or the side that pays the funding. A breakout can be decided by a fixed rule alone, with no model asked.
A grid
A grid sets levels evenly across a range: buys under the price, and over it sells that give back only what is held. Each round trip inside the range pays one step less its fees, and nothing else; the grid earns from the market going back and forth, not from its direction.
Its risk is the price leaving the range. A buying grid out of its range holds what it bought and waits; a selling or neutral one should close everything past a few percent beyond, since the losing side keeps growing.
A range
Two entries rest just inside the borders of the last hours' range, a buy at the bottom and a sell at the top, the first filled cancelling the other. The position leaves at a share of the range's width or at a stop beyond the border. It is made for quiet markets, and stays out of a strong trend.
One strategy, or one a day
Locker Protocol Agentic plays one strategy a day with its daily pilot: eighteen are shipped, and you can write your own from any of them by changing figures, each read against its bounds. Putting another in use never touches a position already held.
The pilot can also choose each day among two to five, by a table of the market's regime or by the model in one brief call, and play the day whole under the one chosen. Whatever it plays, the local limits and the mandate still refuse what they refuse.
Measure first
A strategy is a hypothesis until it has been measured on past candles and funding, with the real fees. Measure it from several starting days rather than one: the same grid can look excellent from one day and poor from the next. The measure asks nothing of the model and places nothing.
This page names no return. A figure measured from one starting day says little, and a past that went well says nothing of the next month. Measure, play it in simulation, then fund an account with only what you are ready to lose.