Methodology
How we validate,and how you check it.
The proof page shows the numbers. This page shows how they were measured, defines every term in plain English, and points you at the raw data so you can reproduce them yourself.
A number you can't reproduce isn't proof. So everything below is built to be checked.
Train forward, test once
The validationpipeline.
The whole method fits in one picture. Train on a window that is frozen the moment testing starts, run forward onto data the model never saw, and measure once.
Development data, the never-seen line, the same curve carrying on.
Train, then freeze
The model learns on a fixed historical window. Once testing begins, that window is frozen. No going back to adjust it after seeing how the test turned out.
Walk forward
It then runs on the data that comes next in time, the out-of-sample slice, exactly as it would trade live, with no knowledge of what happens after each bar.
Measure once
The result is read a single time, after costs, through the shipped exit. First look, only look. That single number is what we publish.
No jargon left unexplained
Every term,plainly.
The words behind the numbers, and why each choice is the honest one. The long version on win rate: why win rate is a vanity metric.
Out-of-sample
Results are measured on price data the model had never seen and was never tuned against: a separate slice of history, held back on purpose.
Why it's honest · An in-sample number only shows how well something memorised the past. It always looks good and almost never survives the future.
Walk-forward
We train on a fixed window, then test on the data that comes right after it, moving forward through time the way you'd actually trade. The future never leaks into the training.
Why it's honest · You can only ever trade on what has already happened. Walk-forward stops the classic cheat of using tomorrow to explain today.
Holdout
For the index rules, a full seven-year block (2015–2021) was set aside and opened exactly once, at the end, to measure, never to tune.
Why it's honest · Opened once means we couldn't peek, adjust and re-test until it looked good. The first look is the only look.
After costs
Costs come off before any validation profit factor is reported: FX at a 0.5-pip spread, indices at real per-trade costs. One exception, stated plainly: the live record is computed from the indicator's own stop and target, so it is gross and reads higher.
Why it's honest · Costs are where most paper edges quietly die, so we take them out first and show what is left.
Through the tool's own exit
The published numbers use the exact stop, target and trailing logic the indicator runs live, not a separate, more flattering research exit.
Why it's honest · It would be easy to validate one exit and ship another. We measure what you actually run.
One position at a time
The tool holds one trade at once, so overlapping signals aren't all taken: about 56% on FX and 68% on indices go untraded. The profit factor applies to the trades actually taken.
Why it's honest · It's how a person trades one account. We report what you could really have taken.
Profit factor
Gross dollars won divided by gross dollars lost. Above 1.0 means the strategy made money after costs. 1.43 means $1.43 came back for every $1.00 lost.
Why it's honest · A tool can win 70% of the time and still lose money. Profit factor captures the whole picture in one number.
R, the risk multiple
One R is what you risked on a trade: the distance from entry to stop. A +2R win makes twice what you risked, a −1R loss is a full stop-out.
Why it's honest · R makes every trade comparable, whatever your account or position size.
Win rate
The share of trades that close in profit. Useful context, but on its own it says nothing about whether a strategy makes money.
Why it's honest · A high win rate with poor risk-reward loses. That's why we lead with profit factor.
The failure mode we avoid
Why this isn'tcurve-fitting.
Curve-fitting tests millions of combinations and keeps the best-looking one. In-sample, a fitted line and a validated line look identical. Past the divider, the fitted one falls apart.
How we stay on the right side of that line.
We don't search for a flattering result. What we publish is out-of-sample, cost-adjusted, framed as uplift over a plain rule-based core, and reported with the losing years included. A backtest optimised until it sparkles is a warning sign, not a selling point.
Honesty is also about limits
What we don't claim.
A method is only as honest as what it refuses to say. Here is what you will never hear from us.
No fixed future profit factor.
The numbers describe a measured period, not a promise about the next one.
Not a validated edge on every market.
PaceAlgo runs on every market and timeframe. The edge is validated on FX majors (5-minute) and ten major indices (daily). Everywhere else it is an honest structure-and-regime read.
Not profitable every year.
On indices, 2018 and 2019 were down. We say so next to the headline, not in a footnote.
Not the research exit.
We only publish the shipped tool-exit, because that is what you run.
Not that you'll match the backtest.
Live results vary by broker, feed, spread and execution. Past performance does not predict future results.
Don't trust us. Check us.
Verify it yourself.
Everything above is only worth something if you can test it without our permission. Three ways to do exactly that.
Recompute the raw data
Every closed forward trade downloads as a CSV: date, symbol, side, module, entry, exit, outcome and R. Recompute profit factor, win rate and drawdown yourself.
Download the trade logWatch it accrue live
Every tracked signal posts the moment it closes in the public Discord, winners and losers, never backfilled. Coverage is still expanding across the validated markets.
Open the DiscordRead the numbers in context
The proof page carries every figure with its sample size, its baseline and the losing years stated alongside. Nothing buried in a footnote.
See the proof