The question
Copy-trading tools show what a "smart money" wallet made, but not what a person copying it would keep. I built Signaldesk to measure that gap for a small retail follower on Solana, after realistic delay, fees and slippage. In a pilot replay, the follower lost money on all three wallets in every scenario with trading costs, including the one wallet that itself gained 29.2 %.
What I built
Signaldesk (signaldesk.tools) is a public tool: paste any Solana wallet, choose a budget, slippage and copy-bot fee, and see what the wallet made next to what a follower would have kept.
- Data pipeline (Python): fetches and caches each wallet's full transaction history, classifies swaps from the wallet's own balance changes, and prices them against one-minute market data.
- Simulation engine: replays every swap as a follower with a fixed budget, position limits, a reaction delay, network and exchange fees, and slippage.
- Live check (JavaScript): a tested port of the same logic that runs in the browser, behind a small serverless proxy that keeps the data-provider key private.
- Design choices: free-tier infrastructure (no running cost beyond the domain), analytics only with no rankings, alerts or trade signals (EU MiCA), and a GDPR privacy notice.
Method
I replayed every swap of three active wallets over September 2026: 6,524 swaps in total. A simulated follower with a 100 EUR budget put 5 % into each copied position, held at most three at once, and paid a 0.05 EUR network fee, a 0.25 % exchange fee and an assumed 1 % slippage on each trade. I ran 64 scenarios per wallet: 4 budgets × 4 delays (0 to 60 seconds) × 4 cost levels.
Data quality was treated as a control, not an afterthought:
- Missing prices are excluded and counted, never estimated; any result with more than 20 % of trades excluded carries a warning.
- A reconciliation test checks that cash, losses and capital still tied up always add back to the starting budget.
- The same inputs always give the same numbers, and the accounting is tested against hand-calculated values.
Findings
On all three wallets, the follower lost money in every one of the 48 scenarios with trading costs. The bridge shows where Wallet C's +29.2 % went at 100 EUR, 30 seconds late, with base costs.
Two mechanisms dominate. Speed: Wallet C began selling 107 of its 169 positions within 30 seconds of buying, so a follower 30 seconds behind usually bought after it was already leaving; in one real trade it sold 15 seconds after buying, up 32.2 %, while the follower bought 25 % higher and lost 4.45 %. Costs: on a 5 EUR position, fees and 1 % slippage need a gain of about 4.6 % just to break even.
Limitations and next step
Three wallets over one month is a pilot, not proof. Prices are one-minute candles, so delays under a minute are coarse; slippage is an assumption because historical pool depth was not available; and 25 % of Wallet C's trades had no price data.
The next step is a pre-registered study fixed before any results exist: the 30 top-ranked wallets on a public leaderboard, chosen on October 2026 data and measured on November 2026 data, with wallets that stop trading or lose money kept in. The headline is fixed in advance as an instant copy with a 1 % copy-bot fee, the best case for a follower, so a loss there holds however fast they are.
Skills shown and links
P&L attribution and variance analysis, scenario and sensitivity analysis, data controls and reconciliation, study design that avoids hindsight bias, and building with regulation in mind (MiCA, GDPR). Built with AI-assisted development; I designed the analysis and can explain every part of it.
- Live tool and one-minute video: signaldesk.tools
- Privacy notice: signaldesk.tools/privacy