One configuration, not three
Tokyo Drift is a structural fork of the Drift VWAP Pullback line. Drift gives Asia, London and New York a complete, independently tuned copy of every setting. Tokyo Drift applies one configuration to whichever sessions are enabled.
Sessions still exist and still matter, but they carry only what is genuinely theirs: the trading window, the end-of-day flatten, the direction toggles, and the trade / loss / reversal budgets. Every filter, entry rule, target, stop and trade-management setting is shared.
The reason is not tidiness. Three independently tuned copies of one strategy is three opportunities to fit the past. A single shared configuration has to earn its keep across every session that is switched on, which is a materially harder test.
Six configurations, and which one to use
Every file below is the same code. Only input defaults differ, so moving between them is a settings change rather than a different strategy. The full parameter list for each is at the bottom of this page.
| file | role | headline |
|---|---|---|
tokyo_drift_v1_5_0.pine |
the published default | 71.49% win, PF 3.458 on the verified TradingView run |
tokyo_drift_v1_5_0_default.pine |
less-optimised alternate | PF 1.16, $13,452, DD $7,011 full sample |
tokyo_drift_v1_5_0_prop.pine |
robustness | profitable in all four periods tested, DD $2,275 over 2.5 years |
tokyo_drift_v1_5_0_prop_balanced.pine |
funded-account evaluations | 65.6% win, PF 1.44, DD $1,341, 11/12 months |
tokyo_drift_v1_5_0_prop_highwin.pine |
highest hit rate | 79.1% win, PF 1.27, DD $1,495, 11/12 months |
tokyo_drift_v1_5_1_xauusd.pine |
spot gold rather than futures | hand-tuned on XAUUSD, no published run |
The default is the hand-tuned Micro Gold configuration. It is the one run live on TradingView and the one the headline figures below come from.
The alternates exist because they were selected on different criteria, and those criteria are worth knowing before picking one. The less-optimised alternate is the untuned development baseline plus the single change that survived a walk-forward — a more conservative starting point for a different instrument, since it has fewer settings fitted to Micro Gold specifically. The robust file was chosen for staying profitable across four independent periods rather than for peak performance, and is documented in the regime-versus-fit study and the prop-evaluation study. The two prop files target funded-account drawdown limits and were selected on the trailing 365 days only, so they carry no out-of-sample validation.
The spot gold file is the odd one out: not another tuning of the Micro Gold configuration but the same strategy pointed at a different instrument. Spot gold and Micro Gold futures share an underlying and very little else that this strategy cares about — a different tick, a different volume profile feeding VWAP, and no session break. Dropping the MGC numbers onto XAUUSD does not reproduce MGC behaviour, so the RSI family was re-tuned by hand until it came close, and the extension filter went on. Thirteen inputs differ from the published default.
Two caveats travel with it, and both matter more than the settings do. It was
tuned by hand over a single 2026-05-24 to 2026-09-03 window with no
walk-forward, so it carries less validation than any other file here. And the
cost model in the script is still the Micro Gold one — $1.20 per side and a
0.10 tick — which is not what a spot-gold broker charges. Put your own spread
into the Properties tab before reading a number off it.
Shipped configuration
Micro Gold, 5-minute, one contract, $2.40 round turn ($1.20 per side).
- Entry — RSI Reset: RSI dips within a 2-bar lookback then recovers through the midline, inside a VWAP trend. RSI 14, reset levels 54 long / 35.5 short.
- Direction gate —
Slope only. VWAP’s slope sets direction; price may be on either side. A trend often turns while price is still on the old side, and requiring the crossing first forfeits the opening of the move. - Filters on — RSI overbought/oversold (59 / 40, 18-bar extreme lookback), MACD Histogram Velocity (18 / 22 / 7), and a 15-point maximum candle range.
- Filters off — 1-hour momentum, Efficiency Ratio, Min VWAP Distance, Max Extension.
- Brackets — ATR Multiple: longs 3.5x ATR against a 2.35x stop, shorts 3.0x against 2.95x.
- Sessions — Asia 7:00 PM–2:00 AM ET (3 trades / 2 losses), London 3:00–10:00 AM (3 / 1), NY 10:30 AM–3:30 PM (3 / 1). One reversal each.
- Reversal Engine — on, full stop-loss only, 3-bar wait, confirming candle required, points bracket with its own trail.
- Breakeven and the primary trail — off.
Results
Verified on TradingView, 5-minute Micro Gold, one contract, $50,000
account, 24 May – 3 September 2026:
| net P&L | +$21,318 (+42.64%) |
| profit factor | 3.458 |
| win rate | 71.49% (168 / 235) |
| max drawdown | $550.40 (0.98%) |
That is a real broker-emulated run rather than an engine figure, and it is the window the configuration was tuned against — so read it as the best case, not the expectation.
The same configuration measured by the engine over longer windows:
| window | trades | net | win rate | PF | max drawdown |
|---|---|---|---|---|---|
| Last 12 months | 716 | $15,120 |
52.7% | 1.30 | $6,935 |
| Full sample 2024-01 → 2026-08 | 1,724 | $8,582 |
47.9% | 1.09 | $13,669 |
The gap between those rows is the honest headline. Over the full sample this configuration runs a 1.09 profit factor with four of eleven quarters positive. The reason is concentration rather than decay: across every configuration tested, the entire 2024-2026 profit falls in two quarters of 2026, and nine of eleven quarters lose money. A volatility explanation for that was tested and does not hold — see the regime-versus-fit study.
Verification
The Python engine twin reproduces TradingView’s fills at 97.6% recall and 97.5% precision entry-by-entry on a paired parameter-and-trade-list export, with entry prices exact on 661 of 663 matched trades. The residual is a data artifact: TradingView prices the continuous gold contract slightly differently from the series the engine reads, and the resulting ~0.35% ATR difference is about half a tick on an ATR-sized bracket.
Every version from v1.1.0 onward was verified to reproduce its predecessor
exactly at default settings before shipping, so the accumulated changes are
additive rather than silently behaviour-altering. That check is also why the
Direction Gate shipped defaulted to its pre-v1.4.0 value and had to be turned
on deliberately for the published default — a version-parity requirement is not
a recommendation.
What was tested and rejected
Eleven research stages and roughly 1,300 configurations were walk-forwarded over eleven quarterly blocks, choosing settings only from data before each cut and scoring on blocks the chooser never saw. Breakeven placement, trailing distance, trailing arm point, proportional profit locks, pullback-triggered stops, reversal brackets, entry delays and slope thresholds all failed to beat simply leaving the strategy alone.
The mechanism is consistent: 98.5% of winning trades pass through the region where any stop-move would fire, so anything that fires clips more winner than it saves loser, and the penalty scales smoothly with how early you intervene. Breakeven and the trail ship off for that reason.
The Direction Gate was the single change that survived.
Known limits
- The edge is two quarters. Nine of eleven quarters lose money on both configurations, and no tested regime variable — volatility included — separates the two profitable ones from the rest. Treat it as episodic and unexplained rather than as a condition that has been identified.
- One instrument, one sample. ATR-scaled brackets will travel to another contract; the point-denominated values — candle-range filter, reversal bracket, trail distances — are gold-sized and will not. The spot-gold file above is the one attempt at a second instrument, and it needed thirteen inputs changed to get close — on the same underlying.
- The published figures are engine-derived. The twin is verified at 97.6% entry-by-entry against a TradingView export, but the table above is not itself a TradingView run.
- Hand-tuning did not help here. The MGC-tuned variant leads only over the last six months, at a lower win rate and a wash on profit factor.