When to switch a bot off — and when it's just a bad week

Both mistakes cost money. Leaving a broken strategy running drains an account slowly. Switching off a working one during an ordinary losing streak locks in the loss and misses the recovery that pays for it. The difference between the two is not a feeling, and it cannot be decided while you are watching the equity drop.

Decide the numbers before you go live

Take them from the backtest, on the longest history you have. Three figures matter: the deepest drawdown, the longest streak of consecutive losing trades, and the longest stretch of days without a new equity high. Write all three down on the day the bot starts, and treat them as the strategy's normal behaviour.

What counts as normal

A strategy whose worst historical run was nine losses in a row will produce nine losses in a row again, and the odds are good that it eventually produces eleven. Real out-of-sample results are usually somewhat worse than the test, so allowing roughly one and a half times the historical figure before you act is reasonable. Inside that band, the correct response is to do nothing.

Most bots are switched off in the last third of their worst historical drawdown, which is exactly the wrong moment.

What counts as broken

Look for a change in behaviour, not a change in the balance:

• Trade frequency shifted. Six trades a week became one, or twenty. The market it was built for has changed shape.
• Holding time shifted. Positions that closed within a day are now open for a week. Usually the exit logic is not being reached.
• The average loss grew but the average win did not. Slippage, spread, or stops being hit before targets.
• Costs changed. A broker widened spreads or raised swap and the edge was thin to begin with.
• The drawdown is deeper than anything in the test. Not slightly. Clearly beyond it.

Any two of those together are enough to stop, and stopping means moving to demo with the same settings, not deleting the bot.

Reduce before you stop

Turning size down by half is a decision you can reverse cheaply. It keeps the strategy in the market while you gather two or three more weeks of evidence, and it costs a fraction of what a full exit costs if the streak was ordinary. Full stops should be reserved for behaviour that has genuinely changed.

Keep a one-line log

Every Friday, write down the week's result, the number of trades, and anything unusual. It takes a minute. After two months you have a record of what normal looks like for this specific bot on this specific account, and the question stops being a judgement call made under pressure.

The uncomfortable part

Sometimes the answer is that the strategy was never as good as the backtest suggested, and the live account is simply showing you the truth. That is worth accepting early, at a small size, rather than late at a large one. The purpose of the three numbers is to make that conversation arithmetic instead of an argument.