Backtests with adult supervision
Most bot marketplaces show a curve that only exists because fees, slippage and survivorship were quietly deleted. anvas replays your graph on candle data, charges the venue's real fees, applies the slippage you specify, and tells you exactly what it did and did not model. Backtesting is the next thing we ship — this is the design we are building to.
Real fee schedules
Maker/taker tiers pulled from the venue you actually trade on.
Candle replay, stated plainly
OHLCV bars down to 1m, with your own slippage assumption.
Overfit score
Every run is graded on how much of its edge survives out of sample.
The three checks we run without being asked
Walk-forward split. Parameters fit on one window, scored on the next — always out of sample.
Regime stress. The same graph replayed through 2020 crash, 2022 chop and 2024 melt-up.
Perturbation sweep. ±10% on every threshold; if edge vanishes, you get a warning, not a green tick.