Proof
We don't ask you to trust us. Here's the evidence.
PaceAlgo's signal detection is validated out-of-sample — measured on data the model was never tuned on, with real costs, and run through the tool's own exit logic. The headline is simple: on the markets where we've proven it, the signals turn a plain rule-based core into something materially better. Below are the numbers, exactly as measured — and the limits, stated in the same place.
Every figure below is the shipped combination: validated entry signals routed through PaceAlgo's own uniform tool-exit, measured out-of-sample with real costs. These are not the research-exit numbers — we don't market the research exit, because it isn't what the tool runs. The honest claim is validated signal detection that beats a plain rule-based core, not a profit-factor promise.
FX majors · 5-minute · out-of-sample
Profit factor · +0.72 vs a plain rule-based core
A plain rule-core loses money on FX 5-minute. PaceAlgo's validated signal selection turns the same market positive — from July 2025 onward, through the tool's own exit, with real spreads (0.5 pip). Both years positive: 1.53 in 2025, 1.30 in 2026.
n = 1,411 trades · GBPUSD · USDJPY · USDCAD · NZDUSD · USDCHF
Worst peak-to-trough: 41.1R drawdown on the way to +256R net
Indices · daily · seven-year holdout
Profit factor · +0.31 vs a break-even core
Deterministic dip-buy detection across 10 major indices, measured on a seven-year holdout (2015–2021) the rules were never fitted to. It lifts a break-even core — but it was not profitable in every year: 2018 and 2019 were down. We state that rather than bury it.
n = 528 trades · long-only · through the tool's own exit
Worst peak-to-trough: 38.1R drawdown over the seven-year holdout
Per pair · FX
The 1.43 isn't an average hiding a dud.
Here is every FX pair on its own — profit factor, the plain rule-core it beats, the number of trades, and the worst peak-to-trough drawdown of its running R-total. Every pair clears its own losing baseline; the weakest is shown, not hidden.
| Pair | Profit factor | Plain core | Trades | Max drawdown |
|---|---|---|---|---|
| USD/JPY | 1.73 | 0.65 | 215 | 5.2R |
| USD/CAD | 1.62 | 0.72 | 205 | 8.2R |
| NZD/USD | 1.59 | 0.64 | 227 | 8.9R |
| USD/CHF | 1.29 | 0.73 | 588 | 34.2R |
| GBP/USD | 1.21 | 0.82 | 176 | 14.5R |
| All five | 1.43 | 0.71 | 1,411 | 41.1R |
How to read drawdown. Each trade risks one R (the distance to its stop); there's no position sizing in the figures. Max drawdown is the deepest the running R-total ever fell from a high — so the pooled FX record dipped 41.1R below its peak at the worst moment on its way to +256R net. At a fixed 1% risk per trade, that peak-to-trough is roughly a 41% equity dip: real, and stated here rather than buried. Measured on the same out-of-sample windows, the same costs, and the same tool-exit as the headline profit factor above.
Per index · Dip-buy
And the indices — including the one that didn't work.
The same dip-buy rules, run on each of the ten indices over the seven-year holdout. Nine were profitable; the Nikkei 225 was a net loss. We show it at the bottom rather than quietly dropping it — a missing row is how averages lie.
| Index | Profit factor | Trades | Max drawdown |
|---|---|---|---|
| DAX 40 | 2.16 | 58 | 3.0R |
| CAC 40 | 1.59 | 61 | 5.8R |
| S&P 500 | 1.51 | 58 | 5.0R |
| Euro Stoxx 50 | 1.42 | 49 | 7.6R |
| Russell 2000 | 1.33 | 27 | 3.1R |
| FTSE 100 | 1.23 | 46 | 7.2R |
| Hang Seng | 1.18 | 34 | 5.1R |
| Dow 30 | 1.16 | 63 | 7.4R |
| Nasdaq 100 | 1.09 | 80 | 8.1R |
| Nikkei 225loss | 0.88 | 52 | 7.5R |
| All ten | 1.31 | 528 | 38.1R |
Holdout 2015–2021, long-only, through the tool's own exit, after costs. Drawdown in R, same basis as the FX table above. The pooled 1.31 already carries the Nikkei loss and the down years (2018, 2019) inside it — it's the honest number, not a survivor.
Coverage
We tell you exactly where the edge is — and where it isn't.
Most tools claim to work everywhere. That's the tell. Here is the honest map: validated signal detection on specific markets, an analysis tool everywhere else, and a tool that stays quiet when there's nothing worth trading.
Signal detection validated out-of-sample, measured through the tool's own exit.
- FX majors5-minute
- Major indicesdaily · long
GBPUSD · USDJPY · USDCAD · NZDUSD · USDCHF and 10 major indices. No profit promise — see the numbers above.
Everywhere else, PaceAlgo is a market-structure & regime tool — not an edge claim.
- Other FX pairs & timeframes
- Metals, stocks, crypto
Clean levels, structure and risk — with no claim we've proven an edge there.
When the setup isn't there, it says so — and stays quiet.
Most indicators only tell you when to trade. We also tell you when not to.
The best trade is often no trade. Selectivity is the point, not a bug.
A number you can't reproduce isn't proof. So we show the method, the costs, and the years that didn't work.
Methodology
What these numbers actually mean.
The short version is below. For the full method — every term in plain English, and how to reproduce the numbers yourself — see the methodology page.
Out-of-sample
Performance is measured on data the model never trained on, and was never tuned against. In-sample numbers are excluded — they only tell you how well something memorized the past.
Walk-forward / holdout
FX signals come from a walk-forward trained model tested on July 2025 onward. The index rules are measured on a seven-year holdout (2015–2021) they were never fitted to. Fresh data, opened once.
Costs included
Transaction costs are subtracted before any profit factor is reported (FX: 0.5 pip; indices: real per-trade costs). Returns shown are after costs, not gross.
Measured through the tool's own exit
Crucially, these figures use PaceAlgo's uniform tool-exit — the exact exit the tool runs live — not a separate, more flattering research exit. What you see measured is what the tool actually does.
Why this isn't curve-fitting
Curve-fitting is what happens when you test millions of parameter combinations and keep the one that looks best on history. It always produces an impressive backtest — and it almost always fails live. We don't search for a flattering result. The figures above are out-of-sample, cost-adjusted, framed as the uplift over a plain core, and reported with the years that didn't work included.
Read this carefully
Honest limits.
- The tool uses a transparent, uniform exit — not the exact exit from our research validation. The numbers above are the shipped combination, not a separate research figure.
- On indices, the strategy was not profitable in every year — 2018 and 2019 were down. There is no “profitable every year” claim.
- The tool trades one position at a time, so overlapping signals aren't all taken (FX ~56%, indices ~68% of detected signals are not traded). The profit factor applies to the subset actually traded.
- Backtests are descriptive, not a promise. Past performance is not indicative of future results, and live results vary by broker, feed, spread, and execution.
- The edge is validated on specific assets and timeframes. On every other market, PaceAlgo is an honest analysis layer — not an edge claim. Trading involves substantial risk of loss; nothing here is financial advice.
Forward track record
The real test runs forward.
Out-of-sample backtests show what survived validation. The forward record shows what happens next — published as it accrues, not curated after the fact. It's still an early sample: read it as a live receipt building in real time, not a proven long-run number yet.
As of Jul 2026 · live at pacealgo.com/proof. Past performance does not predict future returns. Trading involves risk of loss. Not financial advice.
Every closed trade above is posted the moment it closes in our public Discord — winners and losers, with the result, timestamped by Discord itself. You can watch the record build in real time; nothing here is backfilled or curated. To keep the edge fair, entries are shown to members only — the public feed carries the honest outcome, not the tradeable signal. Verify it in Discord →
Prefer raw data? Download every closed trade as CSV → and recompute the profit factor, win rate and drawdown yourself.