The arithmetic is already there. Bitsgap's H1 2026 Crypto Bot Performance Report, published August 7, ranks win rate and return across four bot architectures operating in a market that spent more months trending than ranging. The data exposes a 23-point spread between the least and most adaptive system, and a nearly 9x divergence between mean and median return on the top performer.
The Adaptive Spread
Win rate climbed in a strict gradient across bot types: GRID at 35.14%, spot DCA at 40.95%, DCA Futures at 50.92%, COMBO at 57.85%. DCA Futures and COMBO also closed positions roughly 3.7–5x faster than GRID and outperformed it on every return measure in the dataset.
The ordering is not accidental. A grid profits from price oscillating inside a range; when a market trends persistently in one direction, the strategy keeps buying into a move that never reverses back through its band, and a system built for chop starts absorbing a one-way loss instead.
Mean vs. Median: The Distribution Problem
COMBO's headline average return of 15.77% sits nearly nine times above its own median of 1.76%. A handful of large winners pull the mean upward. The median — the bot in the exact middle of the distribution — describes what a typical closed instance actually earned.
Bitsgap separates the two figures on purpose. When average and median diverge this widely, the average documents outlier-driven, leverage-amplified outcomes rather than base-rate expectation. Collapsing them into one headline number would misrepresent what most traders experienced.
A second limitation sits inside the win rate itself. Any closed bot ending above zero counts as a win, whether the close landed at +0.01% or +40%. A high win rate confirms frequent profit-taking, not meaningful profit-taking. Median return has to be read alongside it.
Execution Cost Remains Outside the Frame
None of the reported figures isolate execution cost. The price a bot's order targets and the price it actually fills at are not always the same number, and that gap erodes realized return on every trade, not only the losers. The report makes no slippage adjustment.
Verdict
The data supports a constrained conclusion. In H1 2026, adaptive position-sizing architectures — DCA Futures and COMBO — outperformed range-bound GRID by a wide margin, with execution speed compounding the edge. But the 9x gap between COMBO's mean and median is the single most important number in the file. The typical bot earned 1.76%. Outliers earned the rest. Risk-adjusted deployment sizes for the median, not the average.