Here's an uncomfortable thing to publish as a company that runs a copy service: copying a profitable trader usually doesn't make you money. We know because we spent weeks building a system to do exactly that — and then killed it, because the data was unambiguous. This is that study, in full, because the reasoning is the whole point of how we run the desk.
The obvious version of copy-trading is: find wallets with great track records, mirror their trades, share the upside. We built the pipeline to test it properly — pulling the on-chain history of consistently profitable accounts, then simulating what would have happened if we'd followed each of their trades a moment later, at a realistic size, paying realistic costs.
If the naive version worked, we'd have shipped it. It didn't. And the way it failed is the useful part.
1. A high win rate wasn't a real edge — it was just the price. When we bucketed profitable traders by entry price, their win rate matched that price almost perfectly: enter at 0.64, win about 64% of the time. That's not skill beating the market; that's exactly what the market already implied. A trader can look brilliant and simply be buying likely outcomes at fair value — which transfers no advantage to whoever copies them.
2. Their real profit lived in things you can't copy. The accounts that did genuinely profit made their money from mechanics a follower has no access to: maker rebates for providing liquidity, size that moves at prices you'll never get, and speed inside fast-moving markets. By the time a copier sees the trade and mirrors it — later, smaller, as a taker — that edge is gone. You're copying the visible trade, not the invisible advantage that made it work.
3. Live reality was worse than the simulation. When we tried real fills, the books were too thin to get in at the prices the copy thesis needed — a run of attempted entries simply didn't fill at anything usable. Even the simulation's thin edge didn't survive contact with a real order book. That gap — backtest to live — is the same one that kills most strategies.
The mistake in most copy-trading — the one we nearly made — is asking "is this trader profitable?" That's the wrong question. The right one is: "does this trader's edge still exist after you copy it — later, at your size, paying your costs?" Very often the honest answer is no, and no amount of an impressive-looking track record changes it.
A profitable trader is not the same thing as a copyable edge. Confusing the two is the entire business model of most copy-trading platforms.
So we made it a rule: before building any copy strategy, the first test is whether the source actually beats its benchmark in a way a follower can capture — one honest question, answered before a line of code. It's the cheapest kill in the book, and it saves you from the expensive ones.
This is exactly why we don't run a leaderboard. Glasshouse doesn't mirror strangers whose edge might be rebates, size or speed you can't replicate. We copy our own systematic desk — strategies where the edge is the signal we publish, validated out-of-sample, promoted through shadow → paper → live gates, with the losers killed in public before they ever touch capital. When the edge is the signal itself, copying it transfers the edge. When the edge is somewhere you can't see, copying the trade transfers nothing.
We published this because it argues against a lazier, easier product we could have sold. We'd rather run the harder one honestly.
Follow the research. Every kill, revival and live update — published as it happens, free.
📘 Free: get our plain-English field guide — the 7 checks before you trust anyone with your crypto.
Join the Telegram → · get the email notes · or talk to us about copying the desk.