When to stop live trading bot and trading bot performance monitoring searches converge on one idea: you need pre-written triggers that force a pause before emotions take over.

Strong triggers

Drawdown beyond the band implied by research and sizing (Position sizing)
Live slippage worse than modeled costs (Cost drag)
Operational failures: repeated API errors, data gaps, partial fill anomalies

Drift versus research

If live trading vs backtest performance gap widens without a clear operational explanation, treat it as a research failure signal, not noise (Why live fails).

Coordinate with validation evidence

Kiploks verdicts help set expectations before live risk (Verdicts).

Write triggers as code, not vibes

The point of a kill switch is to remove discretion in the worst moments. Define thresholds in advance: max daily loss, max weekly drift versus expected distribution, max consecutive losing trades beyond a band that your research supports. If you cannot state the rule in one sentence, it is not operational.

Cooldowns and re-enable rules

Stopping is only half the problem. You also need a re-entry policy: what evidence must exist before you turn the bot back on? Often the correct answer is "new research cycle completed," not "I feel better."

Correlation with manual intervention

If you frequently override the bot mid-session, your live track record is not comparable to the automated backtest. Either automate the intervention rules or accept that you are running a hybrid discretionary system (When is a strategy ready to deploy?).