Is Tokyo Drift's 2024 losing streak a curve fit or a different market? Neither, it turns out

Tokyo Drift’s MGC configuration was hand-tuned during 2026 and forward-tests well. Backtested over the full 2024-01 to 2026-08 sample it returns a profit factor of 1.09. Two readings of that gap are possible and they have opposite implications:

• Curve fit. The settings were selected while looking at recent data, so they describe 2026 and nothing else.
• Regime shift. MGC in 2024 was a different instrument, and no configuration works across both.

They are indistinguishable if you only ever test the tuned config. So three earlier default sets were run on the same bars — all frozen before 2025-26 existed, none ever tuned on this window. Every run below is MGC 5-minute, $1.20/side commission, 1 tick slippage, single contract.

The tuning is not why 2024 loses

A config that never saw 2025-26 loses 2024 just as hard as the one tuned on 2026. Per-quarter the two move together almost exactly:

v1.0.0 is the control that makes this argument work. It is a different entry model on identical bars, and it is anti-correlated — which rules out the +0.954 being an artifact of two configs merely sharing a market. The quarter drives the result, not the settings.

The market genuinely did change:

Average daily range went from $21.55 to $152 — seven times — while price more than doubled.

The volatility explanation looked obvious and is wrong

The natural conclusion is a volatility floor: below some minimum range the bracket cannot clear costs. Splitting the eleven quarters at 0.10% ATR appears to confirm it decisively.

It does not survive contact. Remove the two best quarters from the high-volatility group:

Both are statistically indistinguishable from zero, and the tuned config moves the wrong way. Two further checks agree: 2026Q1 was the highest-volatility quarter in the sample (0.191% ATR) and returned only +$8.88/trade on the Default and −$20.74 on the tuned config; and within the high-volatility quarters, the correlation between ATR and return is negative (−0.257 and −0.514). There is no dose-response, and there is no floor.

What is actually there: two quarters

The whole 2.5-year profit is two quarters, and this holds for both configurations. Nine of eleven quarters lose money.

Those two were gold breaking down from a parabolic top — Q2 closed −15.1% — followed by a very wide-range Q3. Both directions paid in both (Q2: +$8.03/tr long, +$56.05 short; Q3: +$79.37 long, +$68.41 short), so it does not reduce to a short bias. No regime variable tested here separates them from the rest of the sample. The honest description is that the edge is episodic and unexplained, not that a condition for it has been found.

Full per-quarter detail

v1.1.0 is shown for completeness but is not an independent observation: its defaults resolve to the untuned set, so it differs from the Default column only by the one gate setting discussed below.

Which config ships

Across four windows the untuned defaults win three, with roughly half the drawdown:

The tuned config leads only in the last six months, where the profit factor is a wash and the win rate is lower. That is the signature of a configuration fitted to a recent window rather than a better one — a separate question from whether 2024 was a fit artifact, and it lands the other way.

One deliberate deviation from “untuned”

The published Default is not quite the raw pre-release values. It ships Direction Gate = Slope only rather than Side + Slope.

The gate is the one change out of eleven research stages and roughly 1,300 configurations that survived a walk-forward. It defaulted to the old value in v1.4.0 and v1.5.0 only so that each version would reproduce its predecessor exactly at default settings — a release check, not a recommendation. Those raw values are a backward-compatibility baseline, not a curated config, and shipping them unexamined would have published a default that ignores the only validated finding the research produced.

Six of the twenty-two tuned parameters are inert in both configs, incidentally — erPeriod, erThreshold, momThreshold, maxExtensionStdDevMult and vwapDistAtrMult all sit on filters switched off in each. The effective difference between the two configurations is five settings, not twenty-two.

By session

Run in isolation over the full sample, on the Default config:

These do not sum to the combined figure — trade budgets and reversal slots interact when sessions run together. Worth noting that London is positive here at PF 1.04, where under the MGC-tuned config it runs −$5.31/trade at PF 0.91.

What this does not say

One instrument, one timeframe, 2.5 years, and a Python engine twin rather than TradingView itself — the twin matches TV entry-for-entry at 97.6% on a paired fixture, but it is not a broker fill. Nothing here tests MES, MNQ or MBT, where the same logic may sit in a different relationship to the market. And two profitable quarters out of eleven is a small enough base that the next one is genuinely uncertain in both directions: this data cannot distinguish an edge that arrived in 2026 from an episode that has already passed.