Before this site existed, we spent a year building and forward-testing automated strategies with paper money — and, unusually, publishing the failures. Every claim below is something we measured, not something we read.
Two of our bots won 85% and 83% of their trades and still lost money in twelve days. A handful of large losses erased dozens of small wins. The number that matters is profit factor — gross wins divided by gross losses — together with the ratio of average win to average loss. Any leaderboard that sorts by win rate is sorting by the wrong column.
We built a 1-minute scalper properly: volatility filter, ATR stops, circuit breakers. Two independent engines put it at roughly −99% over six months. The autopsy: round-trip cost was ~70% of the entire stop distance, and fees alone were 54% of the loss. It was deployed live anyway, against a failing gate, to test one question — whether live fills behave differently from replayed candles. They did not. It was retired at 46 trades.
Our best strategy returned +11% in the bear year it was designed in. Tested in three unseen regimes — a +2,500% bull, a second bear, a recovery — it lost in all three. Nothing about the rules changed; only the weather did. When you see a spectacular 90-day record, the first question is what the market did during those 90 days.
We optimised a strategy over 64 parameter sets on 70% of history, then ran the winner on the unseen 30%: in-sample +19.8% with a 100% win rate became +2.7% with a 50% win rate. A 100% win rate in a backtest is the fingerprint of luck-fitting. Check your own backtest →
Same strategy at 1%, 2% and 3% risk per trade: +6.7%, +5.5%, −10%. Monotonically worse. Risk systems are non-linear — bigger bets trip the protections that were producing the edge in the first place. "Same edge, bigger size" is false.
Across six years of data, buying and holding — and buying on a fixed schedule — beat every strategy we built, by a wide margin. That result is not inspiring and cannot be sold as a course, which is roughly why nobody sells it to you.
A reel making the rounds advises: when a coin rallies, don't enter immediately — wait 24 hours, because it usually drops back. It is a falsifiable claim, so we measured it over 6.9 years and 3,792 rally events across 30 pairs, with the rules written before looking.
The observation is true. After a daily rally of 15% or more, the price is lower 24 hours later in 57% of cases, median −1.4%.
The advice built on it is false. The mean 24-hour move is +1.5%, because the 43% that don't dip run hard — and waiting skips exactly those. Head to head, entering immediately beat waiting in 8 of 12 comparisons across thresholds and holding periods; waiting won one.
| Rally threshold | Events | Next 24h median | Next 24h mean | Lower 24h later |
|---|---|---|---|---|
| ≥ 10% | 2,376 | −0.72% | +0.92% | 55% |
| ≥ 15% | 954 | −1.41% | +1.50% | 57% |
| ≥ 20% | 462 | −1.76% | +2.26% | 57% |
And the part neither side mentions: both entries have a negative median return at every horizon we tested. Buying after a rally is a fat-tailed lottery; moving the entry by a day does not repair a negative median.
This is the win-rate trap again, in public: "it drops 57% of the time" describes frequency, and profit is frequency times size. If you want to check a claim like this yourself, the same three questions do most of the work: how often, by how much, and what does the median say rather than the average.
A recurring argument in crypto videos: this token burns supply, so it must go up. We tested it on the coins we hold data for, from September 2020 to August 2026.
The deflationary example usually given is BNB, and on the face of it the claim looks strong: BNB returned +2,322% against BTC's +503% — roughly 300 percentage points of outperformance.
Then look at what sits next to it in the same table:
| Coin | Total return | vs BTC | Supply policy |
|---|---|---|---|
| DOGE | +2,551% | +340% | permanently inflationary, no cap |
| SOL | +2,417% | +318% | inflationary, decreasing schedule |
| BNB | +2,322% | +302% | deflationary, quarterly burns |
| LINK | −5% | −84% | fixed cap |
| DOT | −81% | −97% | inflationary |
The best performer in the sample is the most inflationary coin in crypto. Capped-supply coins fill the bottom. Whatever separated the winners from the losers, it was not the supply schedule.
There is a second problem with the BNB case specifically: almost all of its outperformance happened in one year. Against BTC by calendar year it was −119pp, +1,196pp, +12pp, −127pp, +12pp, +29pp, −4pp. Remove 2021 and the "deflation premium" disappears. A single year doing all the work is the signature of a narrative fitted to a price move, not a mechanism driving one.
None of this says burns are pointless — reducing supply is real. It says supply mechanics are swamped by everything else, and that "it's deflationary" tells you almost nothing about what happens next.
We are running a live experiment on copy trading → · public calls we are tracking →
Independent analysis of publicly available Binance copy-trading data. Not affiliated with Binance. Not financial advice. All figures come from Binance's public API at crawl time.