How a signal is graded

Every signal is graded after the fact, losers included, by rules fixed before it fired. Three measurements come out of it: was the direction right, how far did price run for it and against it, and did it beat a random entry.

Direction at fixed horizons

At 1h, 4h, 12h and 24h after the signal, DexiScan checks whether price moved the way the signal said. A move only counts when it is bigger than 0.25 times that coin's own volatility over the horizon. Smaller moves are left unscored as noise, so a coin that barely moved is neither a free win nor a free loss. Where a coin's volatility is not known yet, a flat 2% band stands in.

Upside and drawdown

For 48 hours after a signal, checkpoints record how far price ran in the signal's favour (MFE, max favourable excursion) and the worst drawdown it went through first (MAE, max adverse excursion). MFE shows what was on the table. MAE shows how much heat you had to sit through to get it.

Edge over a random entry

The strictest test replays one fixed trading rule on every signal: enter at the signal, exit by the rule. The same rule then runs on random entries on the same coin and side. The difference is measured in the coin's own volatility (σ), so +2% on BTC and +15% on a small coin can be compared. A signal family beats random only when the 95% interval of that difference clears zero. The random benchmark needs about nine days, so the newest episodes wait as pending.

Episodes, not alerts

A signal that gets re-posted stronger, 6 wallets then 8 then 10, is one episode entered at the first alert. Grading each re-post would grade the same trade three times. An episode ends at the first exit event of its family. For an accumulation signal that is the wallets leaving (INVALIDATED), an opposite wave on the same coin, or a time cap.

Too few to read

Under 30 calls or under 10 separate days, no rate is quoted at all, only the count. Under 100 calls a rate is shown but marked thin.

What grading leaves out

Grading measures the signal, not a trader. Fees, funding, slippage and size are not in a hit rate. The paper portfolios on the dashboard trade every signal at snapshot marks to show what a mechanical follower would have made, and they are labelled SIMULATED.

All of it, per signal type, is on the track record, with a CSV.