Copy-tradingKilled

An upgraded copy-trading bot, with three fixes bolted on

0 of 5 gates

of the pre-set pass tests cleared, on just 2 trades, far too few to judge

In plain English

An upgraded copy-trading bot made only two trades in the test window, far too few to prove anything, so we can’t call it a win.

What was claimed

Three fixes bolted onto the killed copy-trading bot would turn it into a winner.

The bar we locked

30+ trades, beats simply holding, no single trade dominating

What the data said

only 2 trades, all five checks failed

Why it's dead

The v2 fixes (soft-trail, a gentler trailing exit; maker-maker, fee-saving passive orders on both entry and exit; and a regime gate, only trading in favourable market conditions) were tested in a 7-day window that produced only 2 trades, far below the locked n≥30 floor. Insufficient sample; the class isn't permanently retired, but v2 as specified is dead.

The detail

The pre-registered side-by-side comparison (§5 of its own spec) fired 2026-05-17. All five locked gates fail: n=2 (≪30, far below the minimum trade count), beats-HODL-by-$10 false (it did not beat simply holding by the pre-set $10 margin), concentration 75.1% (one of the two trades was three-quarters of the result), down-regime n=0 (zero trades in falling markets, so that check couldn't even run). Direction-of-evidence (n=2, not verdict-bearing) had v2 ahead of v1, but anti-p-hacking rules (no hunting for a friendlier reading after the result is in) forbid rescuing a kill on it.

Kill date: 2026-05-17
Sample: n=2 (7-day window)
Method: Pre-registered live test
Verdict: insufficient sample