Methodology

How we build, judge, and report our trading bots.

Most signal services show you the wins. We publish how our bots actually perform after fees — including the losses. This page is the long version: what each bot does, how we decide when its results are trustworthy enough to show, and why a single number like win rate can mislead you. No demo curves, no cherry-picked screenshots.

The five bots

Five rule-based bots trade independently, each on a distinct, named strategy family. Here is the plain-English idea behind each one and the market conditions it is built for. We do not publish the specific thresholds, lookback windows, or parameter values — those are the part that has to stay private — but the concept is no secret.

OP-1 · Hurst-gated Mean Reversion

Mean reversionRegime-gatedHurst exponent

Buys into calm, range-bound markets that have drifted too far from their average — and exits as price reverts back.

It watches for calm, range-bound markets where price has wandered unusually far from its own recent average, then positions for a drift back toward that average. It deliberately stays out when the market looks like it is trending rather than ranging, because a stretched price in a strong trend is not the same opportunity as a stretched price in a quiet range.

OP-2 · Multi-MA Trend

Trend-followingMulti-timeframeHan–Zhou–Zhu

Follows established trends, entering only when momentum lines up across short and long timeframes together.

It rides moves that are already established, entering only when shorter- and longer-horizon momentum point the same way. Requiring that agreement keeps it out of conflicting, choppy stretches where a trend has not really formed, and focuses it on markets that are committing to a direction.

OP-3 · Donchian Breakout

BreakoutTrailing exitTurtle system

Catches markets breaking out of their recent range, with a trailing exit that lets winners run and cuts losers short.

It treats a decisive push beyond the edge of a market's recent trading range as the possible start of a new move. A trailing exit lets a genuine move keep running while stepping aside quickly if the break stalls, so it is built for markets transitioning out of quiet consolidation into a fresh expansion.

OP-4 · OU Mean Reversion

Mean reversionStatisticalOrnstein–Uhlenbeck

Trades statistically stretched moves back toward fair value — and only when the math says they're likely to snap back fast.

It models how far price has stretched from a statistical sense of fair value, and how quickly past stretches have tended to snap back. It acts only when a return toward value looks both likely and reasonably prompt, which suits temporary dislocations in pairs that are otherwise well-behaved.

OP-5 · Keltner Breakout

BreakoutVolatility-basedKeltner channel

Trades volatility-driven breakouts, reacting earlier than range-based methods when a calm market starts to move.

It keys off volatility starting to expand, reacting when a previously quiet market begins to move with conviction. Because it responds to that change in behaviour rather than waiting for a fixed price level to give way, it tends to engage around the moment a calm market wakes up.

How we judge a bot

We gate results on sample size

A win rate from a handful of trades is noise, not evidence, so we hold results back until there are enough trades to mean something. Under 20 closed trades, the win rate is hidden entirely. From 20 to 99 trades it is shown but flagged as low-confidence. At 100 or more trades it is shown clean. Whatever you see is always the real, after-fees result for the trades a bot has actually closed — never a projection.

We never show win rate on its own

Win rate is the most quoted number in this industry and the easiest to hide behind, so it never travels alone. It is always shown alongside the figures that give it meaning:

Profit factor. Win rate ignores size; profit factor weighs everything won against everything lost, so a high win rate that is undone by a few brutal losers still shows up as unprofitable.
Average win. A bot can win often and still go nowhere if its wins are tiny, so we show the typical size of a winning trade rather than just how often it wins.
Average loss. The same logic cuts the other way: frequent small wins can be erased by occasional large losses, so the typical size of a losing trade is shown right next to the wins.
Expectancy. Expectancy folds how often a bot wins together with how much it wins or loses into the single number that says whether the edge is actually positive per trade.

Read together, these tell you whether a bot makes money — something win rate alone can never tell you.

What “paper” means

Right now the bots trade on paper. That means every trade is simulated against live market prices, with fees and costs applied — but no real money changes hands, and there is no live track record yet. Paper trading is how a strategy earns its place before it risks anything real: it has to clear the sample gate above, in the open, where you can watch it.

We think this is the honest way to start. Rather than launch with a polished history we cannot back up, we are letting the record build in public. As real, after-fees results accrue — the good and the bad — you will see them here.

The short version

We would rather show you an honest, unfinished record than a flattering, unverifiable one. The gates, the after-fees reporting, and the losses-in-plain-sight are all there for the same reason: so that when our results are worth trusting, you have every reason to trust them.

For the full risk, paper-trading, and non-custodial disclosures, see the Disclosures page.