Corrections

What was wrong

This tool is for someone holding dollars who needs euros. The euros you receive are the dollars you have divided by the EUR/USD rate — so a falling rate is good for you, and a rising one is bad. Converting now beats waiting exactly when the rate goes on to rise.

Our scoring code tested the opposite. It marked a "convert" call correct whenever the rate subsequently fell — which is precisely when waiting would have bought more euros. The error came from reading "convert while the dollar is strong", which is sound advice about today's level, as though it were a forecast of further dollar strength. Those are not the same claim, and treating them as the same flips the sign of the answer.

What it affected

Everything downstream of that test: the accuracy percentages on the old track record page, the per-timeframe figures on the homepage, the accuracy labels in the newsletter, and the model weightings the system learned — because the learner was rewarding whichever models were best at calling the value-destroying side.

Publishing was halted on 5 August 2026, the day the error was found.

What we found after fixing it

We rebuilt the scoring, made every model declare what its signal means, and re-ran ten years of history. The honest result is much smaller than what was previously claimed.

Measured properly — in euros gained versus simply converting when the money arrives — the surviving signal is thin, and almost all of it traces to a single macro episode (the dollar rally after the 2024 US election). Strip that one episode out and the remainder loses money. Across 2026 so far, it lost money.

We also worked out how much evidence it would take to prove an edge of the size we appear to have: roughly 900 independent 20-day periods, which is on the order of 70 years of data. An effect that small cannot be demonstrated within the life of a product like this.

What we will claim from now on

No accuracy percentages, and no backtested performance claims. A backtest is not a track record. If a track record returns to this site it will be a forward-looking ledger: every call from a stated start date, with the realised euro outcome against simply converting on receipt, updated automatically and including the losers.

Treat what this site shows as a dashboard of macro indicators, not as a prediction of where the rate is going. For most people converting a regular income, a fixed schedule is a reasonable default, and the interest earned on dollars while you wait is likely to matter more than the timing.

Why publish this

The error was ours and it ran for months. Anyone who read the old accuracy figures was given a number that pointed the wrong way. Correcting it in public is the minimum owed, and the method that caught it — checking a headline claim against the arithmetic of what actually lands in your account — is the more useful thing to take away.