Every performance number you see on Tradevo comes from the same testing process. This page explains exactly how we measure an algo before it is allowed onto the Marketplace — and, just as importantly, what we throw away. Most platforms show you a back-tested curve and stop. We'd rather show you our work.
We test against roughly a decade of daily price history (about 10.9 years for Bitcoin, 10 for Ethereum, fewer for newer tokens) pulled from our own trading venue's candles. The data is pinned and checksummed — frozen to a fixed snapshot — so a test we run today produces the same answer next month. Nobody gets to quietly re-fetch friendlier numbers.
That decade is deliberately brutal: the 2018 bear market (Bitcoin down ~84%), the 2020 crash, the 2021 double-top, the 2022 collapse (down ~77%), and the ongoing 2025–26 bear. An algo has to behave sanely in every one of those regimes — not just the friendly years.
A backtest is a simulation of how an algo would have traded in the past. Ours are run net of trading fees and slippage — the real cost of getting in and out, including the price impact of larger trades on thinner pools. The number you see is after those costs, not a frictionless fantasy.
We also fixed the most common way backtests cheat: no look-ahead. Our engine can only act on a price after it has printed, and fills at the next day's open — never at a closing price the algo couldn't actually have traded at. That gap is small per day but is frequently the entire "edge" of a naive strategy, so we removed it and re-ran everything on the honest engine.
Before testing, we write down what we expect an idea to do and why — its mechanism, the markets it should win and lose in, the parameters it's allowed to use. Then we test it. This keeps us honest: it stops us from running thousands of variations and crowning whichever one looked best by accident. Existing algos get re-entered as fresh hypotheses with no special privileges.
We split history into three parts so an algo can't simply memorize the past:
We also walk forward through time — repeatedly training on one window and testing on the next — and report only the stitched-together out-of-sample result, never the in-sample fit.
Test enough strategies and one will look brilliant by pure chance. We correct for that. The deflated Sharpe ratio takes the algo's risk-adjusted return and discounts it based on how many candidates we tried — the more variations tested, the higher the bar the winner has to clear to prove it isn't just the luckiest draw.
We also demand a parameter plateau, not an isolated peak. If nudging an algo's settings by ±25% causes its performance to fall apart, that result was a fluke and we reject it. We only trust strategies that keep working across a whole neighborhood of nearby settings — and we prefer ensembles (a vote across several settings) over any single hand-picked number.
Finally, an algo has to clear real benchmarks: it must beat simply holding cash (USDC) and stack up on a risk-adjusted basis against simply holding Bitcoin. "Lost less than the market" isn't automatically good if cash would have done better.
In our most recent testing rounds we ran 283 candidate strategies through this gauntlet. 36 passed. The rest were killed. Strategies that looked great in one window but collapsed when a single setting changed were thrown out as overfit. Ideas with a popular following but no real evidence — day-of-week timing, halving-calendar bets, sentiment-as-a-timer, momentum strategies that die once you charge realistic costs — were rejected outright.
This also applies to algos already on the platform. 13 previously listed algos were delisted after re-testing on the honest, no-look-ahead engine — their old numbers didn't survive. Removing things that don't hold up is part of the process, not an exception to it.
A backtest is a hypothesis about the past. The only honest test of the future is the future. So before a validated algo is listed it is replayed over a historical window, and the top picks get a small real-money self-test. Once tracked, we record its actual performance forward from the date tracking began — the "since" date shown on each listing (engine v2) — daily, against real prices, and never revised. Over time you're looking at how it has really done, not just how it once tested. Experimental algos carry a NOT BACKTESTED label and show no performance history.
Backtested and simulated results are hypothetical. They show how a strategy would have performed on historical data and do not represent actual trading. Hypothetical results have inherent limitations — among them, they are prepared with the benefit of hindsight and cannot account for every real-world cost or market condition.
Past performance does not guarantee future results. Crypto assets are volatile and you can lose your entire allocation. Nothing on this page is investment advice. Tradevo is non-custodial and not SEC-registered; we operate under the SEC Covered User Interface Provider safe harbor (effective April 2026).