RESEARCHED HARD.
SHIPPED SPARINGLY.
WITHDRAWN WHEN WRONG.
Seven strategies are available. Two are on by default. Every one of them carries the numbers it actually produced — profit factor, worst drawdown, average win against average loss, and the year-by-year breakdown that shows which years it lost. Four more were shipped and later withdrawn — two for broken validations, two that were measured correctly and simply never earned their place. All four are documented on this page rather than deleted from it.
THE BAR IT HAS TO
CLEAR.
Backtests are trivially easy to fake and even easier to fool yourself with. The process below is designed to make a strategy fail — and most of them do. Only what survives every stage gets shipped, and two things that survived it once still didn't survive a second look.
| Stage | What we do | What kills a candidate |
|---|---|---|
| 1. Hypothesis | A written market thesis first — a specific inefficiency and why it should persist | "The parameters look good" is not a thesis. Curve-fits with no story are rejected here. |
| 2. In-sample build | Rules and parameters developed on one period only | Anything needing more than a handful of parameters to work at all |
| 3. Out-of-sample test | Run untouched on years of data the strategy has never seen | Edge that evaporates on unseen data — the single most common failure |
| 4. Cross-pair check | Tested across 12 currency pairs; pairs where it doesn't work are excluded, not force-fitted | Working on exactly one pair, which is almost always coincidence |
| 5. Permutation sweep | Parameters perturbed to confirm the edge sits on a plateau, not a spike | Results that collapse when a setting moves one notch — a fitted artefact |
| 6. Causality audit | Every indicator read from the last closed bar, with the forming bar dropped, and the simulation checked for any figure that could not have been known yet | Lookahead. This is what caught two of the four strategies we withdrew, and it is now a stage rather than an assumption. |
| 7. Significance test | Compared against random entries held for the same length of time, and resampled by whole trading days to get a confidence range | A result that a coin toss could have produced. "Better than random" is not the same as proven. |
| 8. Reality anchor | Engine output reconciled against genuine broker reports, with real spreads and costs, plus punitive spread stress tests where execution matters | Edges that only exist in a frictionless simulation |
Most candidates are rejected, and four that passed were later withdrawn. A high rejection rate isn't a sign of poor research — it's the whole point. Harder still is pulling a strategy you already shipped: in July 2026 we retired four. Two had validations contaminated by lookahead errors. The other two were honestly measured and simply never earned their place, which is the less dramatic and more common reason to cut something. All four are on this page.
ALL SEVEN,
WITH THEIR NUMBERS.
Everything measured over Feb 2023 – Jul 2026 unless stated otherwise, at the settings that actually ship, and re-derived in July 2026. Profit factor above 1.0 means the strategy made money over the test period; it is not a return, not a forecast, and not a promise.
| Strategy | Character | Profit factor | Status |
|---|---|---|---|
| Mean Reversion | Fades over-extension, banks early, high win rate | 1.73 | Core — on by default |
| Trend Follower | Joins a confirmed trend on a pullback | 1.54 | Core — on by default · demotion reversed |
| Aggressive Survivable | The two above, at materially higher risk per trade | 1.66 | Opt-in · Pro and above · replaces the two above |
| Swing Rider | Same entries, held whole for the multi-day move | 1.27 | Opt-in · Pro and above |
| Gap Fade | Fades the Sunday reopen gap back to Friday's close | 2.82 in pips | Opt-in · needs the engine running all weekend |
| Reversal Capture | Fades a late-session over-extension after the NY close | ~2.3 out-of-sample | Opt-in · demo it first |
| Stretch Fade · new | Rests a limit order beyond an over-extended move and waits | 1.14 | Opt-in · experimental, demo only |
| Momentum Rejoin | Retired July 2026 — never validated, and re-testing moved it further from proof (p 0.102 → 0.158). | Retired | |
| News Momentum | Retired July 2026 — best variant 1.05, indistinguishable from breakeven. | Retired | |
| Regime Breakout | Withdrawn July 2026 — lookahead error. Causal signal: 0.59 out-of-sample, 0.40 live. | Retired | |
| Session Fade | Withdrawn July 2026 — same class of error. 1.98 became 0.77 once fixed. | Retired | |
BUILT TO DISAGREE
WITH EACH OTHER.
Running several strategies that all lose in the same conditions isn't diversification — it's leverage in disguise. This library is deliberately assembled from systems that earn in different market regimes, and their per-year figures show it: the years one of them loses are usually years another one's best.
Mean Reversion earns in calm, ranging markets, one small win at a time. Trend Follower and Swing Rider need the calm to break. Gap Fade only exists at the Sunday reopen, Reversal Capture only wakes up after the New York close, and Stretch Fade sits on its hands until a pair has stretched a long way and then waits to be come to. Their weekly returns are close to uncorrelated, which is why a combination produces a smoother curve than any single component.
You are not obliged to run any of them but the first. Turn strategies on and off from either device, instantly. The engine's exposure caps then govern the whole portfolio jointly, so several enabled strategies still cannot stack into one oversized currency bet.
Aggressive Survivable is the Mean Reversion and Trend Follower logic at higher risk — it is designed to replace them, not to run on top of them, and the console says so. Swing Rider takes the same entries as Trend Follower but holds them whole, so running both means two positions on the same idea. Neither combination is blocked, because your account is yours — but neither is diversification, and you should know that before you switch it on.
Mean Reversion
Core · on by defaultFlagshipThe strategy the whole engine is anchored to. When price stretches unusually far from its own recent average it leans against the move — buying dips and selling spikes, entering patiently behind price rather than chasing it. It banks the bulk of each winner early, lets a small remainder run, protects the trade once it is modestly ahead, and sits behind a deliberately wide safety stop. It earns its keep in the long, calm middle of most trading weeks.
That figure was never measured on this strategy. On 25 July we re-derived every profit factor and mapped four of them onto the wrong strategies; the number printed here belonged to a parameter set the engine has never run, and Mean Reversion's real result was published under Aggressive Survivable's name. Nothing about the engine changed and no strategy was altered — the measurement was mislabelled. The four affected figures were all understated. What we changed so it cannot recur →
When it earns
Ranging and gently trending markets with normal liquidity. Its bank-early exits capture the snap-back before it fades, which is why it accumulates a very high win rate rather than waiting for home runs.
When it struggles
Sharp one-way moves and news shocks. Read the two numbers together: a 94% win rate with an average win about an eighth of the average loss means a single losing trade gives back what roughly eight winners earned. The high win rate is not the edge — surviving the losses is. Expect long stretches of small gains punctuated by one that hurts.
SUITS: traders who want frequent, small, steady activity and can genuinely sit through a losing month · GUARDED BY: cooldown after stop-outs, per-pair cap, spread guard (fades need tight execution), the drawdown throttle
Trend Follower
On by default againThe trend rider. It waits for a genuine change of trend to confirm, then joins it on a pullback rather than at the turn, entering behind price in two steps. It banks most of each winner early, takes a slice more if the move extends, and leaves a runner out behind a wide stop.
From 25 to 28 July this page said Trend Follower scored 1.00 — exactly breakeven, that we claimed no edge for it, and that we would not suggest starting here. We also switched it off on every account running it.
All of that was wrong. The 1.00 was measured on a completely different parameter set — one that takes the opposite side of the same signal, and which this engine has never run. On the settings that actually ship, Trend Follower measures 1.54, is positive in all four years, and has the shallowest drawdown of the four. The seasonality claim we made at the same time (that summer runs at 0.97) came from the same mismeasurement and has been withdrawn rather than restated — it was never tested on this strategy.
The demotion is reversed and it is on by default again. If you deliberately switched it off, check the toggle after updating — the setting cannot distinguish "never touched it" from "turned it off on purpose", so it may come back on.
When it earns
Sustained macro moves and rate-differential runs. It is the natural complement to Mean Reversion — where reversion wants calm, this wants the calm to break — and its two strongest years are the two most recent, at 1.79 and 2.07.
When it struggles
Chop. A confirmed trend change that immediately reverses gives it a full-size loss for a partial-size win, and 2024 was its weakest year at 1.23. Its payoff is as lopsided as its siblings: about £50 average win against £409 average loss, so one bad trade undoes eight good ones. It trades a good deal less often than Mean Reversion, so expect quieter stretches.
SUITS: traders who want a genuine complement to reversion rather than a second helping of it · GUARDED BY: weekend gap guard (runners hold overnight), exposure caps, news guard
Aggressive Survivable
Opt-in · Pro and above · explicit enableA high-octane compounding profile: the Mean Reversion and Trend Follower logic running together at a materially larger risk per trade than standard. It is designed to replace those two rather than run alongside them.
That figure was measured on a single parameter set which is, in fact, the Mean Reversion engine on its own — not this composite. Corrected to 1.66 across 4,007 positions. Like the other three, the error understated the result.
When it earns
The same conditions as its two components, amplified — and it remains the most consistent of the four, positive in every year tested, with 2025 at 2.02.
When it struggles
Everywhere the others struggle, but louder. That 93% win rate is real and is not the point: at £48 average win against £401 average loss, the strategy spends most of its time collecting small wins and occasionally hands a large one back. It also needs your account guards widened to match — Equity Guardian around 25% and a daily loss limit around 12% — or they will halt it during swings that are normal for it.
SUITS: experienced traders sizing deliberately, on money whose loss they can absorb · GUARDED BY: every guard, plus the drawdown throttle that scales its risk down automatically while underwater
The old brief quoted a 3,000-path simulation: "median +75% per six months, 0.2% chance of halving the account." We withdrew it on 25 July because it assumed profit factors of 1.8 and 1.45 for its two components and "neither reproduces — they measure 1.12 and 0.94". On shipped settings those components are 1.73 and 1.54 — the second of them now above the 1.45 the study assumed — so that reason was itself a product of the mismeasurement, and the study had defined its own inputs correctly all along.
Re-run on 29 July over 50,000 paths, every claim holds: median +81% per six months against the +75% claimed, 0.12% chance of halving against 0.2%, drawdowns of 21% typically and 33% at the bad end against the "routine 20%, occasional 35%" originally stated.
We are still not putting those odds back, for three reasons. They describe six months — over three and a half years the chance of halving roughly doubles and the median drawdown grows from 21% to 35%. The actual 2023–26 path drew down 51%, breaching the very line the model prices at one-in-450, so either that path was extraordinarily unlucky or the simulation understates the tail; with one real sample you cannot tell, and the flattering reading is not the one to bet on. And the compounding figures come out at CAGRs we would not put in front of anyone, whatever the arithmetic behind them.
| Risk per trade | Chance of halving | Median, 6 months | Typical drawdown |
|---|---|---|---|
| 1.0% | 0.0% | +20% | 6% |
| 2.0% | 0.0% | +43% | 12% |
| 3.5% — shipped | 0.1% | +81% | 21% |
| 5.0% | 1.1% | +125% | 30% |
50,000 simulated six-month paths, resampled in blocks from the measured daily results, with an allowance applied for backtest optimism. Backtested and simulated — not a forecast.
That table is what we would rather you took from the study. Going from the shipped setting to 5% multiplies your chance of halving the account by about ten and adds roughly a quarter to the median. The dial matters more than anything else you can change here — which is a far more useful thing to know than a reassuring headline probability.
A higher-risk option that's enabled by default isn't a choice — it's a trap. Aggressive Survivable is off until you deliberately turn it on, sits behind the Pro tier, and shows its full risk character before you enable it. We'd rather sell fewer upgrades than have a customer discover their risk setting during a drawdown.
Swing Rider
Opt-in · Pro and aboveThe patient one. It takes the same trend-change entries as Trend Follower, but holds the entire position for the big move: no early profit-taking, a volatility-scaled stop, protection locked in once the trade is clearly ahead, and from there a trail that follows the best price the trade has seen and never gives ground. Typical holds run days to weeks.
The 1.10 this page used to show was never measured on this strategy; like the other three it came from the wrong parameter set. Measured properly, the previous exit scored 1.24.
Separately, we changed the exit. Because Swing Rider takes identical entries to Trend Follower, we could test exit policy on its own across 1,343 positions — same signals, only the exit varying. Adding a trailing stop took it from 1.24 to 1.27, with 14% less drawdown and about 8% more profit. That part is a real improvement, not a restatement.
The same study answered a question the page had only asserted: remove the protective stop move entirely and the strategy collapses to breakeven with three to seven times the drawdown. Holding a whole position through a swing without protection is not a strategy.
When it earns
Sustained multi-day directional phases. It wins less often than anything else here — 72% against the low nineties — but keeps far more of each winner: £121 average win against a £249 average loss, where the bank-early strategies run nearer £30 against £300. That is the entire point of holding the position whole.
When it struggles
Range-bound weeks, and any period where a multi-day thesis is invalidated overnight — 2024 measured 0.92 and remains a losing year even after the new exit. Trend Follower scores higher (1.54) with under half the drawdown. The two took identical entries until 31 July 2026, when USDJPY was dropped from Trend and deliberately kept here — it loses Trend money and makes Swing money, so the comparison is no longer quite like-for-like. Swing Rider earns its place by making about a fifth more money in absolute terms, for close to twice the drawdown. Choose it deliberately, and judge it over quarters, not weeks.
SUITS: traders who prefer swing holds to scalping and can hold through overnight risk · GUARDED BY: weekend gap guard (essential here), margin floor, exposure caps · Trades 11 pairs
Gap Fade
Opt-in · operationally demandingEvent-driven, and the cleanest thesis in the library. When the market reopens after the weekend, price has usually gapped from the Friday close — and across 1,572 gaps on 12 pairs, 92% of gaps of 8 pips or more fully retraced to that close within 48 hours. This strategy fades each fresh gap exactly once: it trades against the gap with the target set at the Friday-close level, a stop one full gap beyond entry, and a 24-hour give-up exit if the fill never comes.
When it earns
Almost every weekend. The reopen gap is a genuine liquidity artefact rather than a pattern in noise, which is why this is the highest profit factor in the library and why it survived a punitive reopen-spread stress test. Since 0.9.154 it also asks your broker whether it has already traded a given gap, so a restart cannot double up.
When it struggles
The risk is not visible in the win rate. Gaps appear on many pairs at the same reopen — typically 7 at once, up to 16. At 1% risk per trade, a bad weekend therefore puts far more than 1% at stake, and the worst weekend in the sample cost 6.1% of equity at that sizing. Size for the weekend, not for the trade.
OPERATIONALLY DEMANDING: it only fires at the Sunday reopen, so the engine must be running across the weekend — a VPS or the hosted engine is ideal · NOTE: the ~3.5 figure quoted for this strategy previously is the same result expressed in risk units rather than pips; both are real, they are different scales
Reversal Capture
Opt-in · demo it firstThe night owl. After the New York close the market thins out, and a pair that has stretched a long way from its recent range often runs out of participants and snaps back. Reversal Capture waits for the turn to actually begin — price pulling a few pips off the extreme — then fades it, with a tight stop just beyond the extreme, a fixed target, a breakeven lock and a time-stop. It hunts a narrow window each night, because that is where the edge measurably is.
When it earns
Thin, post-close liquidity where an over-extended move can't sustain itself. We broke the results down hour by hour and kept only the hours that actually pay; earlier evening hours were tested and dropped. Because those hours span rollover, we re-ran it charging triple and quintuple the normal spread — it stayed profitable at both.
When it struggles
Genuine late-session breakouts that keep running: the fade gets stopped for a small, controlled loss. It trades rarely and the per-trade edge is modest, so it earns through consistency rather than fireworks. The stop is small enough that spread matters, which is exactly why it is restricted to eight pairs.
ORIGIN: proposed by a SentryQ trader, reshaped and validated before it shipped · GUARDED BY: every guard, plus the rollover window and spread guards that cover the hours it trades into
Stretch Fade · new
Opt-in · experimental · demo onlyThe patient contrarian, and the strongest backtest in the app after Reversal Capture. When a pair has pulled a long way from its own recent average it waits — and instead of chasing, it rests an order even further out and lets the move come to it. If price never reaches the order, nothing is traded and nothing is lost: the order simply expires. When it does fill, the position is entered at a better price than the one that triggered the idea. Volatility-scaled stop, fixed target, protection only once well ahead, and entries deliberately spaced so one large move cannot spawn a cluster of trades that all fail together.
What makes it interesting
Unlike its siblings it does not rely on a 90%-plus win rate and a lopsided payoff — 54% wins with a positive profit factor is a very different, and more robust-looking, shape. It was positive in all three test windows including the most recent one, its confidence range sits entirely above 1.0, and ten of twelve pairs make money. It also survived every artefact test we aimed at it.
The one risk no backtest can settle
It enters on resting limit orders, and whether those orders really fill at the modelled prices in a live market is not something a simulation can prove. The study deliberately charged a pessimistic fill — price had to trade well past the limit before it counted — precisely because filling on a wick-touch flatters a limit strategy. Live fills are the only test. That is what demo trading it is for; the P&L is not the point.
⚠ DO NOT RUN THIS ON A LIVE ACCOUNT until it has a demo record of its own. Trades 12 pairs, both directions · A note on its design: the identical signal taken at market instead of on a resting order is a loser — the edge is smaller than the spread. That is the single most important fact about this strategy.
TUNE WHAT YOU SHOULD.
NOT WHAT YOU SHOULDN'T.
Every parameter ships at the value that survived validation. Pro and above can override the ones that are genuinely yours to own — while the signal logic itself stays locked, so tuning can't quietly destroy the tested edge. Blank a field and it returns to the validated default.
| Setting | Who can change it |
|---|---|
| Strategy on / off | All plans, desktop or mobile, instantly |
| Risk per trade | All plans — the most important dial you own |
| Stop distances & take-profit levels | Pro and above |
| Order behaviour & entry offsets | Pro and above |
| Per-strategy trading windows | Pro and above |
| All twenty-six risk guards | All plans — protection is never a paid upgrade |
| Which pairs may trade | All plans — bench a pair from the safety panel any time |
| Reset to validated defaults | All plans — one click, per strategy or for all of them |
| Signal-generation logic | Locked. Not editable by anyone, including us, outside a versioned release |
THE ENGINE MARKS
ITS OWN HOMEWORK.
Each strategy carries the win rate and expectancy it produced in validation. The engine tracks live results against that envelope on a rolling basis and puts the verdict on the card: on track, WATCH or DEGRADED — with the actual numbers, not a vague warning.
A strategy that quietly stops working is the most expensive failure mode in automated trading, because nothing alerts you until the damage is done. Ours is built to say so early, while switching it off is still a cheap decision.
The guards keep their own scorecard too: every entry they refused is replayed against real prices with the strategy's own stop, so you can see whether a block saved you money or cost you some. How the guards work →
THE RESEARCH THAT
FAILED.
Published, not buried. If we only ever showed you what worked, you'd have no way to judge whether our process is rigorous or merely lucky — and no reason to believe the winners weren't cherry-picked.
| Candidate | What we tested | Verdict |
|---|---|---|
| Fibonacci retracement | 144 configurations across retracement levels, confirmation filters, stop placements and pairs | No configuration produced a robust out-of-sample edge. The apparent structure did not survive unseen data. Nothing shipped — the Fibonacci tool on the Trading Desk is there for you to draw with, not because we found an edge in it. |
| Regime Breakout | Shipped, then withdrawn in July 2026 after a live audit | Its validation contained a lookahead error — the simulation read an H4 bar's close at that bar's open, worth a fictitious profit over 2023–26. The causal signal has no edge (out-of-sample 0.59) and every rescue we tested failed. Live results had been telling us the same thing at 0.40. Retired; existing positions were managed out, not dumped. The full post-mortem → |
| Session Fade | Shipped, then withdrawn in July 2026 after the same audit | Same class of error: the study credited a take-profit using the entry bar's own low, which normally occurred before the trade was entered. Removing that one line took the profit factor from 1.98 to 0.77. A rolling walk-forward over Dec 2022–Jul 2026 returns 0.83, it loses in every year, and nothing rescued it — not the target, the stop, the news filter, nor trading the break the opposite way. Retired. |
| Momentum Rejoin | Shipped demo-only, then retired on 29 July 2026 after a re-test | Nothing here was mismeasured — its figures always came from the config it actually ran. It was cut because re-testing made the case weaker: against random entries held for the same length of time it scored p = 0.102 when it shipped and p = 0.158 a month later, moving away from significance as data arrived. Full-span profit factor 1.23; 2026 reads 1.02. The one improvement that survived a selection rule fixed in advance reached 1.28, which does not make an unproven strategy fundable. A candidate that spends a month forward-testing without getting closer to proof is not in progress. |
| News Momentum | Shipped demo-only, then retired on 29 July 2026 | It never had a demonstrated edge and was never going to earn one by waiting: the best out-of-sample variant reached 1.05 — indistinguishable from breakeven — and most sat below 1.0. Its own brief told you not to fund it, which is a fair warning to attach to a strategy and a poor answer to why it was on the list at all. News trading is the hardest style to validate, so the honest reading is that we do not know how to trade it, not that this attempt was unlucky. Retired. |
| A hedge for late-session reversals | The idea that a losing fade could be rescued by opening the opposite side | Refuted. The hedge cost more than it saved once spread was charged at the hours involved. What survived the study was the plain retrace-fade that became Reversal Capture — the negative result is what pointed at the positive one. |
| Breakeven moves and partial profits on Momentum Rejoin | Two standard "improvements" any sensible reviewer would suggest | Both measured worse. A breakeven move roughly halved the profit factor; partials held the profit factor but cut around 45% of the return. They are documented as absent-on-purpose so a future tidy-up doesn't reintroduce them. |
Publishing 144 failed Fibonacci tests means openly admitting that months of work produced nothing sellable. Admitting that our own re-measurement was botched — as we do immediately below — costs considerably more. We do it because the only meaningful signal of research quality is what a firm is willing to reject, including its own work, and because you should be deeply suspicious of any trading product whose research has apparently never once failed.
On 25 July 2026 we re-ran every strategy at its shipped settings and republished the results, including where they were worse. That exercise was itself broken. The tool doing the measuring worked from a hand-copied list of settings that had drifted from the engine, and four results were attached to the wrong strategies.
Mean Reversion's genuine result was printed under Aggressive Survivable's name. The number we published as Trend Follower's came from a setup that takes the opposite side of the same signal — one this engine has never run — and on that basis we told you the strategy was breakeven, withdrew our claim of an edge, and switched it off on every account running it.
Nothing about the engine changed and no strategy was altered. The measurement was mislabelled. Corrected on 28 July:
| Strategy | We published | Actually measures | Positions |
|---|---|---|---|
| Mean Reversion | 1.11 | 1.73 | 2,727 |
| Trend Follower | 1.00 — "no edge claimed" | 1.54 | 1,280 |
| Aggressive Survivable | 1.38 | 1.66 | 4,007 |
| Swing Rider | 1.10 | 1.24, now 1.27 with a new exit | 1,343 |
Every one of the four was understated. We are aware that a correction which happens to flatter us is the least credible kind, which is why the whole method is described here rather than summarised: the figures were traced back to the exact rows of an old parameter sweep they came from, matched to the pound, and Mean Reversion was then re-measured a second time using a separately written script over a longer window, which returned 1.81 against the 1.73 quoted above.
What we changed so it cannot happen again. The measuring tool no longer keeps its own copy of the settings — it reads them out of the engine's source directly, and refuses to produce a figure at all if it cannot. A separate check compares the live settings against a fingerprint recorded when the current figures were measured, and must pass before any number goes into a brief, onto this site, or into marketing copy. It caught its first real drift the day it was written.
The rest of the library reproduced correctly: Gap Fade (2.82 in pips) and Reversal Capture (~2.3) were measured by their own separate replicas and are unaffected. Stretch Fade (1.14) was published as experimental from the start and remains so. News Momentum and Momentum Rejoin were also measured correctly — and were retired on 29 July because correct measurement is exactly what showed they had not earned a place.
Three things we publish that we previously didn't. First, maximum drawdown — the peak-to-trough loss a strategy put its own equity curve through. A return you couldn't sit through isn't a return you'd have collected. Second, average win against average loss: most of these systems win over 90% of the time, which sounds remarkable until you see the average win is roughly an eighth of the average loss. Third, the year-by-year breakdown, because an average across four years hides the year it lost.
Profit factors and win rates on this page are backtested and simulated results, shown after out-of-sample validation. They describe how each system behaved historically — they are not a forecast, not a promise, and not a return you should expect. Drawdown figures are quoted per lot and scale with your position size. Live results differ, markets change, and strategies degrade. That is precisely why the engine monitors expected-vs-actual and tells you when they diverge. Read our full evidence standard →