Glasshouse Research · July 2026 · 5 min read

Kill criteria: deciding when a strategy dies

Most trading strategies don't get killed — they get abandoned. They run badly for long enough that everyone quietly stops paying attention, the position sizes shrink, and eventually someone notices the bot hasn't fired in six weeks. There was never a decision. There was just drift.

Kill criteria are the fix. A kill criterion is a number you set before a strategy runs live: "if live profit factor drops below 1.0 after 30 closed trades, we stop." The decision is made in advance, with a clear head, before a losing run starts applying pressure. When the criterion fires, it fires automatically — no deliberation, no "but the last three days looked better."

Why pre-commitment is the whole point

The problem with evaluating a struggling strategy in real-time is that you're always inside the losing run. That's exactly the worst moment to decide objectively. The losses feel like a patch of bad luck. The next trade feels like the one that will turn it around. The market has been "weird lately." These aren't excuses — they're genuine pattern-matching from a human brain doing what it's built to do. But they're the same thoughts every trader has right before the strategy compounds the damage.

Pre-committing to a number cuts through this. When the strategy reaches n=30 trades with a profit factor below 1.0, the answer is already decided. The criterion fires on data, not on how the last week felt.

What a kill criterion actually looks like

A useful kill criterion has three parts:

  • A metric — profit factor is the cleanest because it captures the ratio of gross wins to gross losses and doesn't care whether you had 3 trades or 300. Win rate alone is insufficient: a strategy can win 70% of trades and still lose money if the wins are small and the losses are large.
  • A threshold — what counts as "not working." Below 1.0 means the strategy is losing more than it's making. Some desks set it tighter — 1.1 or 1.2 — to catch marginal edge sooner.
  • A minimum sample — below 20–30 trades, the numbers are noise. Killing a strategy after 5 losing trades is just reacting to variance. The minimum sample forces you to distinguish a bad run from a broken edge.

What happens when the criterion fires

The strategy stops. Not paused, not "monitored closely." Stopped and logged. The entry goes into the kill log with the date, the metric at termination, and the reason. That log only ever grows — entries don't get edited or deleted. A kill is not an embarrassment to hide; it's evidence that the process is working.

Revival is possible. If there's a genuine new hypothesis — a different configuration, a new filter that explains the losses — the strategy can come back to paper testing, validated from scratch. A revival also gets logged. The audit trail is permanent.

The five kills on this desk

As of July 2026, five strategies have been killed on data. None were abandoned — each had its criterion fire and was stopped deliberately. The most recent was a mean-reversion strategy that passed its backtest (profit factor ~1.05) but showed PF ~0.66 across 13 paper trades. The backtest edge was thin to begin with; trading costs ate it in live conditions. Stopped and disabled on data. The kills feel like failure. They're the safety system working.

What to ask any service you follow

If you're evaluating a trading service or copy account, the presence of kill criteria is one of the most useful filters you have:

  • Do they have pre-committed kill criteria for each strategy?
  • Are those criteria public — or at least visible to you on request?
  • Have any criteria ever fired? If yes, what happened?
  • If no strategy has ever been killed, is that because they all work — or because the criteria don't exist?

A desk that has never killed a strategy isn't necessarily dishonest — but it's a question worth asking. Killing something is evidence of a process. An unbroken run of survivors is evidence of something, too.

The kill log on this desk is public and append-only. Every strategy that was stopped, every criterion that fired, every revival attempt — logged in the open as it happens.

See the full audit trail → · Join the Telegram for weekly updates · or talk to us.

Important. Glasshouse Research is an educational publication. Nothing here is financial, investment, legal or tax advice, a recommendation, or a solicitation. Backtested and past performance is not a reliable indicator of future results. Trading crypto carries a high risk of loss. Glasshouse is independent and not a licensed financial services provider.