CFD Trading Bot on Gate — what the research actually shows

The tier-A basket: real alpha on real samples

Six instruments where the best configuration beat the instrument's own buy&hold over the forward year on 100+ trades. Long and short — that is why falling markets are represented.

Costs charged in every backtest: Gate CFD spread + commission with a 2× safety margin, pessimistic fills, entry at next bar's open. Full ranking of all 53 instruments (including every tier D/E failure) lives in the research console. These six were measured on the first unified grid; the ATR re-check below is the new standard and is still pending for them.

Live forward book — accruing right now, not a backtest

On 2026-07-17 we froze the best validated configuration for each instrument below and started replaying it forward-only on completed bars with the same engine and the same costs as the research above. Nothing is backfilled: the record starts at zero and earns (or loses) in public.

Paper record, updated every 30 minutes on completed bars only. Zero trades means the strategies are waiting for their entries — trend systems trade rarely by design. Net % is per-trade return on margin summed across the basket, before any live frictions we cannot simulate. PAPER_ONLY: this is evidence being collected, not a performance promise.

Audit log: what the latest re-check changed (15–16 July 2026)

We re-ran part of the battery with per-instrument stops scaled to each market's own volatility (0.5–3× ATR14 instead of one shared percent grid — 160 optimizer jobs), and stress-tested the flagship under doubled costs. Numbers move when methodology improves; we publish both sides.

Why publish this: with honest per-instrument stops, three of four re-checked "winners" died — forex on these CFDs is too efficient for our battery and we say so. The survivors (DAX, gold) are stronger claims than before, not weaker. Standing rule going forward: every new campaign runs on ATR-scaled grids only.

Without optimization vs with our optimization

What buy & hold actually paid over the last 12 months

Real candles, computed 16 July 2026. This is the honest context every bot claim must be read against: the same year rewarded some assets and punished others — and no one knows next year's column in advance. Long/short research exists precisely because of this.

Per-asset data coverage varies (247–365 trading days on our candle store); crypto measured on spot USDT pairs. Buy&hold of a falling market is a real loss a bot can short — that is why the tier-A table above shows falling markets with positive bot years.

What a deposit would have done (that same forward year)

Capital split equally across the six tier-A books, bot trades automatically 24/5. Leverage multiplies both profit and drawdown — we cap public scenarios at 3×.

*One already-elapsed backtest forward year (07.2025–07.2026), scaled to account size — not a forecast. Verified by re-simulation: gold at 1×/2×/3× returned +43/+87/+131% with drawdowns 7.4/14.8/22.2%. Above 3× the drawdown approaches forced-liquidation territory, so we do not show it.