RevertX — The Parameter Dossier
Six pairs, zero tuned parameters, and a throne built on fourteen trades. This dossier asks the uncomfortable questions — and finds that the parameters were never the problem. The problem is that the strategy we simulate is not the strategy that was designed: RevertX has been fighting with one hand tied behind its back since Round 1.
V.1The defendant
Design wait for a confirmed ranging market (ADX < 20 AND flat SMA-50) · then demand FOUR agreements: RSI stretched · price at a Bollinger band edge · Stochastic extreme · %K above/below %D · enter the fade
Designed exit "price returns to the BB midline (the mean) — or ATR stop" · that is the whole point of a mean-reversion trade
Roster config EURUSD GBPUSD USDCHF USDCAD AUDUSD NZDUSD · H1 · params: pure defaults · ATR×1.5 stop, 1:2 target · magic 7701
Round-4 record sealed exam: PF 2.34 · GBPUSD · 14 trades
V.2The starving gauntlet — why so few trades
Every entry must pass two gates and four checks simultaneously. The census, across five years of H1 bars (six pairs are similar; GBPUSD shown):
| condition (alone) | % of bars true |
|---|---|
| ADX says "ranging" | 20.5% |
| SMA-50 flat | 41.1% |
| both gates open (regime) | 10.6% |
| RSI stretched past 30/70 | 26.9% |
| price at a band edge | 42.3% |
| Stochastic extreme | 43.3% |
| everything at once → signal | 0.15% |
One bar in roughly seven hundred. That is ~36 trades per pair per five years — scarcity is not a bug of the parameters, it is the personality of an AND-stack. The checklist that makes each trade high-conviction is the same checklist that starves the sample. Remember this trade-off: selectivity and statistical confidence pull in opposite directions.
V.3The fourteen-trade throne
RevertX's roster seat rests on that Round-4 GBPUSD exam: PF 2.34 across 14 trades. How impressed should we be? We handed a coin-flipper the same 1:2 bracket — a trader with no edge whatsoever — and let 200,000 of them take 14 trades each:
A no-edge trader beats PF 2.34 5.8% of the time — about one coin-flipper in seventeen. The 95th-percentile fluke reaches PF 2.67. Fourteen trades simply cannot tell a modest edge from a lucky streak; compare Sentinel's throne, which rests on ~1,900 trades. RevertX's baseline is a hypothesis, not a fact — and its wobble band in the forward test must be treated as extremely wide.
V.4The exit betrayal — the biggest finding
Now the discovery that reframes the whole dossier. In our simulations, RevertX's median holding time is one hour — a single bar, and ~90% of exits are "signal". Why? Its entry conditions are a one-bar pulse (the Stochastic agreement rarely survives to the next bar), and the engine closes a position the moment the signal lapses. The design says "hold until price returns to the midline" — but the simulation slams the door one bar after entry, every time. We have never actually tested the strategy as designed.
So we built its intended exit and ran it: ATR×1 stop, take-profit at the (moving) BB midline, no signal exit. Same entries, the designed journey:
| pair | roster bracket (as simulated) | designed midline exit | n |
|---|---|---|---|
| GBPUSD | PF 1.76 · +$91 | PF 2.03 · +$227 | 36 |
| USDCAD | PF 0.48 · −$81 | PF 1.07 · +$19 | 41 |
| AUDUSD | PF 0.49 · −$68 | PF 0.99 · −$3 | 32 |
| EURUSD | PF 0.93 · −$14 | PF 0.91 · −$24 | 38 |
| USDCHF | PF 0.35 · −$108 | PF 0.89 · −$22 | 33 |
| NZDUSD | PF 0.50 · −$51 | PF 0.68 · −$61 | 25 |
Under the wrong exit, RevertX loses on five pairs of six. Under its designed exit, four of six rise to breakeven-or-better and GBPUSD strengthens to PF 2.03 over 36 trades — a sample nearly three times the baseline's, pointing the same direction. The strategy was never as bad as our tables said; our harness was mis-fitting its exit. (The engine now knows how to trail on indicator flips, but a take-profit that moves with the midline is a new capability — it goes on the docket.)
V.5The tuning court — Round 4 vindicated
Could different parameters rescue the roster-bracket version? We swept every axis, pooled across all six pairs, tuning half only — then examined survivors out of sample:
| variant | IS (pooled) | OOS (pooled) |
|---|---|---|
| defaults | PF 0.62 | PF 1.08 · +$16 |
| ADX gate 25 (looser) | PF 0.86 | PF 0.96 |
| ADX gate 30 (looser still) | PF 0.93 | PF 0.95 |
| wider band proximity | PF 0.55 | PF 0.95 |
| softer RSI zones 35/65 | PF 0.59 | PF 0.74 |
Loosening any gate buys more trades and loses more money — the gates are not too tight, the harvest beyond them is just thin. No tested variant beats the defaults out of sample. Round 4's "defaults beat tuning" verdict stands: the parameters were never the problem.
V.6Dossier verdict
Parameters: unchanged — every tuning direction fails out of sample; the defaults are the least-bad. But the roster entry is under review for two deeper reasons: ① its exit is mis-simulated — the designed midline take-profit (docket: engine needs a moving-target TP) turns five losing pairs into one winner + four breakevens, so any fair verdict must wait for that capability; ② its evidence base is one lucky-sized sample — GBPUSD is the only pair that earns its keep under either exit (PF 2.03, n=36 designed; PF 1.76, n=36 as-simulated), while the other five pairs contribute noise at best. Recommendation to El: narrow the roster entry to GBPUSD only for the forward campaign, and treat RevertX's forward test as the real Round 1 of its life — with a wobble band wide enough to drive a truck through.
Docket additions: ③ engine support for a moving take-profit level (TP at a per-bar column like bb_mid), ④ RevertX native-exit sealed exam once ③ exists, ⑤ roster-pair narrowing decision (El).
Next dossier: BreakoutGuard — seven pairs on H4, eleven trades of evidence, and the crowd at the door.