Download
App Store
Google Play
How to turn "I think this works" into a number you can actually check — and where most people get stuck.
You have an idea. When this pattern shows up you buy, and you get out 3% lower. You scroll back through the chart and it looks right. Then you put real money behind it, and you have no way of knowing whether you were right or whether you just picked the handful of candles that agreed with you. Quant trading exists to close exactly that gap.
Quant trading is using data and statistics to define your entry and exit rules in advance, then letting those rules decide the trades instead of judging each one in the moment. The rules can be simple (a moving-average crossover, an RSI threshold) or complicated (multi-factor models, machine learning). Complexity is not the point.
The point is that a written rule produces the same answer every time you run it. That is what makes it checkable. You can take those rules, run them over four years of historical data, and get concrete numbers: what the return was, how deep the worst drawdown got, how much of it the fees ate.
These three words get used interchangeably, and mixing them up is the single most common reason a promising idea never becomes something you can run.
| Term | What it is | What it does not include |
|---|---|---|
| Signal | The output of one condition — typically +1 (long), 0 (flat), −1 (short). It answers is something happening right now? | How much to buy, when to cut, what to do if the order fails |
| Strategy | A signal plus position sizing, risk limits and execution rules. This is the thing that can actually be deployed and held accountable. | How it interacts with your other positions |
| Portfolio | Several strategies or symbols combined with a weight allocation across them. | — |
Most people who say "I have a strategy" have a signal. That is not a criticism — a signal is the hard, creative part. But a signal on its own cannot be backtested honestly, because the result depends entirely on the sizing and risk rules you have not written down yet. Two traders with the same signal and different sizing get completely different outcomes.
In Blave Agent this distinction is structural: a signal-based single-symbol strategy is Type A and requires a backtest, a multi-symbol portfolio with a weight vector is Type C and also requires a backtest, and everything else is Type B, which does not.
| Manual trading | Quant trading | |
|---|---|---|
| Entry and exit | Judged in the moment | Written rules — same input, same answer, every time |
| Can it be validated? | Not with the same logic you actually used | Yes — Sharpe, max drawdown, fees paid, all computable |
| When it runs | Only while you are awake and watching | 24/7 once deployed |
| What goes wrong | Discipline breaks under pressure | The rule keeps running after the market it was built for is gone |
Note the last row. Automation does not remove failure modes, it swaps them. A manual trader abandons the plan at the worst moment; an automated strategy keeps executing a plan that stopped working. Both are real, and the second one is quieter, which makes it more dangerous.
Here is a real example from Blave Agent's own backtest engine — btc_sma_cross on BTCUSDT 1h, SMA45/SMA100, 2022–2026. Total return was more than double buy-and-hold. It also had a −42.6% maximum drawdown, and it paid 22% of its starting capital in fees over those four years.
All three numbers matter, and only the first one is the one people quote. A strategy that doubles buy-and-hold while asking you to sit through a 42% drawdown is not obviously better than one that returns less and never scares you out. And a fee bill worth 22% of capital means the trading frequency is not free — a faster version of the same idea can hand its entire edge to the exchange.
The workflow that produces those numbers is always the same five steps: write the idea as explicit rules → backtest it over a period that includes at least one real drawdown → check whether the result survives changing the parameters slightly → deploy it → monitor it. Step three is where most ideas quietly die.
Blave Agent splits "having the idea" from "building the infrastructure". You describe the strategy in conversation — through Telegram, a browser-based VS Code workspace, or SSH if you prefer a terminal — and the agent researches it, writes the code, runs the backtest, and deploys it to your own dedicated server where it runs and monitors itself around the clock.
Two things are worth calling out because they are usually the expensive part of doing this yourself:
Data. Market data comes from Blave's own API — crypto, Taiwan equities and futures — so you are not separately licensing feeds and maintaining ingestion for each one.
Execution. Binance, BingX, OKX and Gate.io are officially supported, with shipped and tested order libraries covering both spot and futures; Taiwan equities route through SinoPac. Other venues the agent wires up on request.
If you do not have an idea yet, the official strategies are free and every one ships with its real backtest. They also have to clear three gates before they are listed: an honest backtest, a Monte Carlo permutation test for statistical significance, and a parameter-robustness check. A strategy that fails those does not get listed.
Five checks. If a strategy fails any of them, the backtest number is not telling you what you think it is.
Can you write the rule as one sentence? "When X is true, do Y, and exit when Z." If you cannot, you have an instinct, not a strategy — and there is nothing to test yet.
Does the backtest window contain a real decline? A strategy that has only ever been tested through a bull market has not been tested. Look for the worst period in your asset's history and make sure it is inside the window.
Are fees in the backtest? The btc_sma_cross example paid 22% of capital in fees over four years. Use a realistic rate (FEE = 0.0005 matches Binance maker/taker at 0.05%) rather than assuming they round away.
How many parameter combinations did you try? Scan a hundred and a few will look significant on luck alone. If you tuned the parameters on all your data and then reported the result on that same data, you are reporting training error.
Could you sit through the worst drawdown in the report? −42.6% on a $1M account is a $426,000 paper loss. If that would make you switch the strategy off, size down until it would not — the drawdown you cannot hold is the one that turns a paper loss into a real one.
It will not find the edge. Backtesting validates an idea you already had; it does not generate one, and a systematic process wrapped around a rule with no edge just loses money more consistently.
It will not make the past a promise. A backtest is a measurement of what already happened under conditions that may not repeat. Market structure changes, and rules built for the old structure keep running anyway.
And it will not remove the human decision. You still choose which drawdown you can live with and how much capital to commit. Those choices sit outside the model, and they decide more outcomes than the model does.
You do not have to arrive with a perfect strategy. Every official Blave Agent strategy is free, ships with its real backtest, and has cleared the three gates above. Deploy one, watch how it behaves through a losing week, and modify it into your own version once you understand what its numbers are telling you.
