GateKilled

Does the trend-farming bot beat doing nothing on fresh days?

Backwards

the bot’s “good day” filter actually picked the worse days, the wrong direction

In plain English

The bot’s own “good day” filter actually picked worse days than the ones it skipped, the filter points the wrong way.

What was claimed

A filter that only lets the bot trade on favourable days will improve its results.

The bar we locked

‘good’ days beat ‘bad’ days by a statistically real margin

What the data said

no real difference, and the “good” days actually did worse

Why it's dead

The farmer performs worse on days that pass its own entry filter than on days that fail it. The gate is anti-predictive. A pre-registered kill fired mechanically on the locked threshold.

The detail

OOS n=53 days (out-of-sample: fresh data the filter was never tuned on). Pre-registered 8-part spec locked the success criterion: Welch p<0.05, mean_pass>mean_fail (a standard two-group statistical test: days the filter passed had to beat days it failed, by a margin too big to be chance). Result: days that passed the filter averaged a net loss while days that failed it averaged ~flat, the wrong direction, Welch p=0.4848.

Kill date: 2026-04-27
Sample: 53 days, fresh data
Method: Pre-registered live test
Verdict: pre-reg kill, gate fired as designed

What it would have done to your money

You put in$10,000

You run it for200 trades

You would have

$8,869

−$1,131

lost (11%)

You started with$10,000

This strategy left you$8,869

Instead of not trading$10,000

This strategy had no real edge, so there is no return to compound. The only honest number is the cost: every trade just pays the 0.06% round-trip fee, compounded here. You choose the amount and the time; the bleed is what trading noise costs.