ArdenAlgoKill zones & sessions
- Kill zones are session windows (London, New York) that many ICT traders treat as the only valid hours.
- This engine runs all 24 hours with no session filter of any kind.
- Session conditioning is a recorded v2 idea, and it is exactly the sort of parameter that invites overfitting.
What the concept says
Kill zones divide the day into windows — typically the London open and the New York open — and argue that moves inside them carry institutional intent while everything else is noise. In equity and FX markets this has an obvious structural basis: those are the hours when the venues that matter are open.
Why v1 skips it
Two reasons, one principled and one statistical. Crypto perpetuals never close, so the structural argument has to be imported rather than assumed. And a session filter is one of the cheapest ways to overfit a small sample: with a few dozen trades and a few candidate windows, some window will always look better in hindsight. Adding it before the engine has passed a basic null test would mean tuning a parameter on noise.
What we would need first
A pooled sample large enough that hour-of-day could be tested as a conditioning variable with a proper null benchmark and a split-half check — the same bar the gate ladder sets for every other conditioning claim. Until then, running around the clock is both the simpler and the more honest choice.
Common questions
Does the engine ignore time completely?
It uses time only for bar boundaries, higher-timeframe aggregation, a per-symbol cooldown and a 48-hour trade expiry. No hour of the day is privileged over another.
Crypto trades 24/7 — do sessions even apply?
That is the open question. Crypto has no closing bell, but it does inherit activity patterns from the markets its participants also trade. Whether those patterns carry information here is untested, and we would rather test it than assume it.
Why is a session filter risky to add?
Because there are only a handful of session definitions and a small sample of trades, so it is trivially easy to find an hour window that flatters past results. That is the textbook shape of an overfitted parameter.