Not indicators. Not patterns. These eight pieces of arithmetic. Everything on this page is illustrated with real numbers from our grading engine on six years of data — including the expensive lessons we learned trading real money ourselves.
Every strategy, no matter how complicated, compresses to one number: how much you make or lose on the average trade, after fees.
If that number is positive, time is on your side and the job becomes sizing and discipline. If it's negative, nothing else on this page can save you — not leverage (it scales the negative), not averaging down (it delays it), not a better win rate (see the next section). Every Tessen grade leads with net expectancy per trade in basis points because it's the honest headline; total return is what marketing leads with.
The most popular bot recipe on the internet — buy the RSI dip, take profit at +1.5%, no stop — graded like this on 2,247 out-of-sample trades through our engine:
It wins three trades out of four and still loses almost everything, because the average loss is 3.4× the average win. A high win rate with a tiny take-profit and no stop doesn't remove risk — it reschedules it into rare, catastrophic losses that arrive after months of confidence-building small wins. That psychological sequencing is why the trap works so well, and why win rate is the single most-marketed number in trading. The verified grade is public — and there are four more like it on the comparison page.
We didn't learn this from a textbook. Our own first live strategy ran a 78% win rate with a 1 : 0.34 risk-reward — flat-to-losing over six years of backtest once we finally tested it honestly, and losing real money live with leverage on top.
Win rate and R:R are two halves of one equation; quoting either alone is meaningless. The break-even win rate for any R:R is:
Read that last one again: a strategy that risks 1 to make 0.3 must win more than 77% of the time before fees just to break even. That's the regime most scalping bots and signal groups quietly operate in. When someone advertises a 90% win rate, the first question is never “how do I get in” — it's “what's the R:R, and what's the expectancy after fees?”
We ran a 1,000-path Monte-Carlo simulation on the real out-of-sample trades of a gate-passing agent, at different leverage levels. The result was blunt: at maximum leverage, every single path ended in ruin — 100% blowup probability on a strategy with a genuine edge. The growth-optimal size was modest, a small fraction of what the exchange would happily let you use.
Leverage doesn't change your edge; it changes your variance — and variance is what kills you, because losses compound asymmetrically: −50% needs +100% to recover. Size from the stop distance and a fixed risk percent (1% is a sane default), never from “how much can I buy.” This is why every Tessen grade includes a responsible-sizing drawdown simulation, and why our risk caps are hard caps.
Tune a strategy long enough on one stretch of history and it will “work” — you've memorized that history's accidents, not learned its structure. The tell: brilliant backtest, mediocre live results, always.
The honest defense is out-of-sample testing: fit on one window, evaluate only on data the strategy has never seen. Every Tessen grade splits history 60/40 — the gate is decided exclusively by the 40% the strategy never touched during design. Most backtesting tools let you optimize and evaluate on the same window, which is exactly how the marketplace of 90%-win-rate strategies gets minted.
Rules of thumb: fewer parameters beat more; a strategy that only works with one magic setting (RSI 14 works, RSI 13 and 15 fail) is memorizing noise; and 30 trades is the bare minimum before a result means anything at all.
A taker fee of 0.05% per side sounds free. It isn't:
High-frequency strategies don't just need an edge — they need an edge bigger than this constant drain, on every trade, forever. When we scanned the classic indicator signals over six years, several had a real pre-fee edge; almost none survived fees. Slippage adds more, and it's worst exactly when you most need the fill — during the volatility spike your stop-loss fires into. Any backtest that ignores fees and slippage isn't optimistic; it's fiction. The Tessen engine models fees on every simulated trade.
Losses and gains are not symmetric, which has two consequences. First: capping per-trade risk isn't optional hygiene, it's the whole game — a string of six 2% losses is recoverable routine; one 40% loss is a different career. Second: drawdown is a budget you spend to earn returns, so two strategies with equal returns and different max drawdowns are not equal — the shallower one lets you size larger and compounds faster. It's why drawdown sits inside every Tessen grade's risk score, with automatic halts (daily loss and max-drawdown) built into every live deployment.
Markets alternate between trending and mean-reverting regimes, and a strategy is implicitly a bet on one of them. Buying dips prints money in a ranging market and bleeds to death in a downtrend; breakout-chasing does the reverse. Most “my bot suddenly broke” stories are really “the regime changed and the bot didn't know.”
You can't reliably predict regime, but you can detect it (trend strength via ADX, realized volatility terciles) and gate your entries with it — our own live systems block counter-trend entries in confirmed trends for exactly this reason, and AgentSpecs support regime filters natively. The subtler point: a backtest that spans only one regime is an overfit in disguise, however many trades it has. Six years of crypto data covers two full cycles — that's why we grade on all of it.
Everything above becomes concrete the first time you grade a strategy you were sure about. It takes a minute and it's free.