Two things are hidden on purpose: which instrument you are trading, and the second half of its price history. Both would let hindsight do the thinking instead of your rules. Build your strategy on the visible slice, then watch how it behaves on the part you never saw — a risk-free way to catch overfitting before real money is involved.
Disclaimer: Trading financial instruments involves risk. Simulated and historical results do not guarantee future performance, and this page is for education only — not financial advice.
To keep results accurate, confirm the basic parameters below before building rules — changing them resets the whole workspace.
Each template is a complete plan — capital, position sizing and rules (entry + exit) all filled in. The symbol is picked at random for you, so a template never tells you what you are trading.
Nothing is locked: every number stays editable. Each card also tells you where the idea breaks down — which matters more than how it looks on history.
A carefully stacked 3-condition entry that looks brilliant on history. Import it, then take it to the Training Ground and watch it fall apart.
Where it breaks: Out-of-sample data. Every extra condition was tuned to one specific stretch of the past — that is overfitting, and the blind test is designed to expose it.
Waits for price to push above the recent high, then rides the session's first real move.
Change any value, then load the template into the editor below.
Where it breaks: Choppy, range-bound sessions. Almost every breakout fades straight back, and at intraday frequency the round-trip costs turn a small loss into a steady bleed.
Buys short-term exhaustion — oversold on a fast RSI — and exits as soon as price normalizes.
Change any value, then load the template into the editor below.
Where it breaks: Trending sessions. Oversold keeps getting more oversold, and mean-reversion takes its worst losses on exactly the days that look most obvious.
A fast/slow average cross compressed onto intraday bars — follows the direction of the last few minutes, flat the moment it turns.
Change any value, then load the template into the editor below.
Where it breaks: Flat, low-volume stretches. The averages cross back and forth with no follow-through, and every whipsaw pays a full round trip in fees.
Choose the symbol you want to study, your starting capital, and how much to risk on each trade. Fees and slippage are real numbers — what you see is what the market would have charged.
Two sentences are enough to start: when to get in, and when to get out. Stack conditions like building blocks — no code, and every rule reads back to you in plain language.
Run the plan over real price history and see what actually happened — the equity curve, the drawdown, and every entry and exit marked on the chart. Judge it by the numbers, not by how convincing the idea sounds.
The stock training ground is where you build a strategy on one part of random historical data, then validate it on data the strategy has never seen. It helps you understand overfitting and develop judgment before touching real money.
In-sample data
The portion of history used to build the strategy.
Out-of-sample data
The unseen portion used to validate — a reliability check against overfitting.
Build on the first slice
You assemble a rule on the in-sample window: buy strength above the 50-day average with a trailing exit.
Commit before peeking
The out-of-sample window is hidden until you lock the strategy — the blind part is the whole point.
Reveal the validation
The strategy is replayed on the unseen slice, and you compare in-sample versus out-of-sample returns.
Read the gap
A small, consistent gap suggests robustness; a sharp drop from in-sample to out-of-sample is the classic overfitting signature.
Pick a market
Choose the stock, ETF or crypto pair you want to study.
Build the rules
Combine entry, exit and position rules as visual blocks — no code required.
Run the test
The engine replays the rules bar by bar over real price history.
Read the result
Review the return, drawdown and equity curve, plus a plain-language assessment of strengths and weaknesses.
In-sample return
How the strategy scored on the data it was built on.
Out-of-sample return
How it scored on unseen data — the more honest number.
Consistency
Whether the two results stay reasonably close, which suggests robustness.
✗ Peeking at the validation window, then adjusting the rules
✓That destroys the blind test. The out-of-sample slice is only informative if the strategy was fixed before it was revealed.
✗ Tuning until the in-sample result is spectacular
✓A perfectly fitted rule is memorizing noise. Simplicity that survives validation beats complexity that only works in-sample.
✗ Treating one passing run as proof the strategy will work live
✓One validation pass is one data point. Live trading adds costs, liquidity, and regime shifts the simulation cannot see.
Use it when
Skip it when
Lock the rules, then reveal
Write the strategy down before the blind window opens. The discipline is the test — the data is just the exam.
Expect some in-sample to out-of-sample decay
A modest drop is normal and healthy. A collapse is the signal to simplify or discard.
Run several random windows
Because the data slice changes each run, surviving multiple different windows is stronger evidence than acing one.
Overfitting happens when a strategy fits the noise of a specific dataset rather than a real pattern. It looks great in-sample but falls apart on new data.
Splitting prevents you from 'cheating' by tuning rules to the very data you test on. The unseen portion gives an honest estimate of robustness.
There is no fixed threshold, but a gap that flips the strategy from strong to weak is a red flag. A modest, consistent edge on both sides is far more believable than a stellar in-sample result that fades on unseen data.
No. It runs simulations on historical price data only, so nothing is at stake except what you learn. That is the whole point: burn your mistakes here instead of in a live account.
It is especially useful for beginners, but the train-then-validate discipline applies at every level. Professional researchers use the same underlying idea to catch strategies that only look good in hindsight.
Start free, scale as you grow. No hidden fees, no surprises.
The full platform with usage limits — no card, no countdown.
Best for
Testing whether an idea survives contact with real data.
No monthly allowance — signup credits only
Deeper history, and your first intraday intervals.
Best for
Swing and intraday strategies on 15- and 30-minute bars, over a decade of context.
Includes 360 credits/month · about 72 runs with review
Every interval we support, and the longest intraday windows.
Best for
High-frequency rules that only make sense on 1- to 5-minute bars.
Includes 1000 credits/month · about 200 runs with review
The training ground uses random historical data for education. Simulated results do not reflect live market conditions and do not guarantee future performance.
Disclaimer: The tools and content on this site are provided for educational purposes only and do not constitute investment advice. Trading and investing involve a substantial risk of loss, and past performance does not guarantee future results.