On this page
Three ways to test a rule
| Backtest | Paper trading | Live, small size | |
|---|---|---|---|
| Data | Historical | Live market, simulated orders | Live market, real orders |
| Speed | Years of data in minutes | Real time | Real time |
| Fills | Assumed by your code | Estimated from live quotes | Real |
| Costs | Whatever you remembered to model | Estimated | Real |
| Can you fool yourself? | Very easily | Somewhat | No |
| Good for | Throwing out bad ideas quickly | Checking a rule survives real timing | Finding out what it actually returns |
They are stages, not alternatives. A backtest can only tell you a rule is bad. Paper trading tells you whether it survives real timing. Only real orders tell you what you would have been filled at.
The data a backtest needs
For tokens that live for hours, candles are not enough. A one-minute candle that opens at 100, touches 40 and closes at 90 does not tell you whether a stop at 60 could have been filled anywhere near 60. You need the trade tape: every trade with its time, size and the pool's state after it. That data comes from the chain itself, decoded transaction by transaction, as on-chain analytics explained describes. You also need the coins that died. A dataset built from tokens that are still listed somewhere has already removed most of the losers.
Seven ways a replay lies
1. It fills you at your trigger price
Code that says "sell when the price falls 40%" records a sale at exactly minus 40%. A real stop is an order sent after the price is seen, into a pool that is falling. New tokens routinely drop 50 to 90 percent within a single minute. In our own forward tests in October 2026, stops set at minus 40% were filled between minus 45% and minus 70%. The same happens in reverse with take-profits: when we replayed earlier picks from stored prices with the take-profit filled at exactly two times, the result was better than anything the same rule produced once it ran live.
2. It trusts stale prices
Many free price sources cache their answers for up to a minute. A replay built on them thinks it knows a price that the market had already left. Sixty seconds is a long time for a token that is ten minutes old.
3. It believes prices from empty pools
A pool with almost nothing in it still quotes a price. Multiply that price by supply and you get a market cap. In our data, one emptied pool with 13 dollars of liquidity implied a market cap of 3.1 trillion dollars, and another showed a "92 times" gain on one cent of liquidity. Both would have been the best trades in any backtest that did not check whether a position could actually be sold. Every price needs a liquidity check beside it before it counts.
4. It forgets you move the price
Buying 1 SOL of a token whose pool holds 30 SOL moves the price against you by several percent, and selling moves it again. A replay that buys and sells at the recorded price assumes your order was not there. The thinner the pool, the bigger the lie.
5. It undercounts costs
Every round trip pays pool or curve fees twice, a network fee twice, usually a priority fee or tip, and sometimes rent for a new token account. Failed transactions cost fees and buy nothing. On small positions held for minutes, these are not rounding errors; they can be the whole edge.
6. It uses information from the future
The subtle one. Filtering on "coins that reached 100K market cap", ranking by a peak that had not happened yet, using a holder count read after the fact, or labelling a wallet as smart using trades it had not yet made. Each puts tomorrow's knowledge into yesterday's decision. If a result looks too good, look for this first.
7. It finds patterns in noise
Try enough rules on one dataset and one of them will look brilliant by chance. With a few hundred trades, most of the profit usually comes from two or three outliers, and removing them reverses the conclusion. A rule tuned until the backtest looks good has learned the dataset, not the market.
A checklist for an honest test
- Decide the rule before looking at results, and write it down.
- Use trade-level data that includes dead tokens.
- Fill at the next trade after your trigger, never at the trigger, and through the pool's real depth for your size.
- Refuse any price without enough liquidity to sell into.
- Charge every cost on both sides, plus a share of failed transactions.
- Use only what was knowable at that moment. Read every input and ask when it became available.
- Hold back the newest third of the data. Tune on the older part, judge once on the newer part.
- Report the median, and the result without the top three trades, beside the average.
- Then paper trade it forward for long enough to see a bad week.
- Then trade it small, and compare real fills with what paper said.
Paper trading against live trading
Paper trading removes the future-knowledge and stale-data problems, because it runs in real time. It does not remove the fill problem. A paper fill is an estimate from a quote, a curve calculation or a pool's reserves; it cannot reproduce latency, a transaction that fails to land, or the price moving between quote and block. Expect live results to be worse than paper results, and treat the size of that gap as the most useful number paper trading gives you.
BLIX has a paper mode for this stage: the same screens as live trading, simulated orders on live Solana coins, with take-profit, stop-loss and trailing stops. The paper trading page says exactly how its prices and costs are estimated and where they can differ from real execution. BLIX does not offer a ready-made backtesting tool, and this guide is not a claim about what any strategy returns.
The short version
A backtest is a filter for bad ideas, not evidence for good ones. The more illiquid and short-lived the asset, the less a replay is worth, and new tokens are as illiquid and short-lived as markets get. Most strategies that look profitable in a memecoin backtest are measuring one of the seven errors above.
More on Solana and on-chain data
Memecoins are extremely risky and most go to zero. Nothing on this page is financial advice.
Frequently asked questions
What is backtesting in crypto?
Running a trading rule over historical price or trade data to see what it would have done. It is useful for rejecting bad ideas quickly; it is weak evidence that a rule will work, especially on illiquid tokens.
What is paper trading?
Placing simulated orders on the live market with no real money. It tests a rule in real time, which avoids hindsight, but its fills are estimates and are usually better than real ones.
Is paper trading the same as backtesting?
No. Backtesting replays the past in minutes. Paper trading runs forward in real time on live prices. A sensible order is backtest, then paper trade, then trade small.
Why do backtests look better than live trading?
Mostly fills and costs: a replay sells at your trigger price with no slippage, ignores your own price impact, and often forgets fees and failed transactions. Hindsight in the data and over-tuning add the rest.
Can you backtest memecoins?
You can, with trade-level data that includes dead tokens, but the results deserve little trust. Prices gap, pools are thin and most tokens vanish, which are exactly the conditions replays model worst.

