Skip to main content

How it works

How does Strategy Arena work?

Strategy Arena Lab judges a trading strategy on a part of the market history kept out of the search, then shows what was measured. The paper arena comes second. No public system receives broker credentials, places an order or holds real capital.

Test a strategy — 0 € See an example report

1

The sealed test

A strategy tuned on a period always looks good on that period. The only fair question is what it does on the part the search did not use.

  1. SealThe most recent 20% of the market history is put aside before anything is tried. An AI working through MCP gets 30% by default, and can set 20% to 50%.
  2. SearchSettings are tried on the remaining 80% only. Every try is counted.
  3. Open onceThe chosen version runs a single time on the sealed part.
  4. CompareThe result is set against Buy & Hold: buying at the start of the sealed part and doing nothing.

2

Four verdicts

Each verdict comes with the numbers behind it: closed trades, result, Buy & Hold over the same bars.

Survives

Every control passes on the sealed period: at least 10 trades, a profit above Buy & Hold, PBO under 50%, Deflated Sharpe of at least 95%, SPA significant, most CPCV paths positive, at least 10 trades in training.

Mirage

A loss on the sealed period, or a clearly negative overfitting control: PBO of at least 50%, or Deflated Sharpe under 50%.

Inconclusive

At least 10 trades and no mirage signal, but the controls cannot decide. A profit at or below Buy & Hold lands here, as does a Deflated Sharpe under 95% or a control that could not be computed.

Not enough trades

Fewer than 10 trades on the sealed period. Too little to judge: try a more active rule or another market.

A script judged as it is, with no search (the online judge), has no selection: PBO and CPCV do not apply, and the other controls decide. In certified mode, a sealed period already opened by the same rule family gives inconclusive at best.

The Lab interface labels them KEEP (survives), REJECT (mirage) and INCONCLUSIVE (the other two). The Survival Audit report uses the same three labels with its own server gates: REJECT on a loss, a drawdown over 30%, a result under Buy & Hold or under an SMA 200 filter; INCONCLUSIVE with too few trades.

A payment never changes a verdict. A bad verdict is still a delivered report.

3

Three places to run it

On the website

The free online judge gives a verdict on a Pine script in your browser. Survival Audit runs it on the server and delivers a private report.

Open the judge →

On your PC

The Lab for Windows and Linux searches variants locally, counts them and judges on a sealed period, from rules you assemble or from your own Pine.

Download the Lab →

In the public arena (secondary)

75 public strategies hold virtual portfolios on timestamped market prices, with no real money. It is not the judge. Gains and losses are public.

Open the dashboard →

4

Read a result without fooling yourself

The +172% that became −37%

A search found +172% on the data that selected it, then −37% on unseen data. The full case.

Read the case →

Strategy Hospital

Why a public strategy is flagged, with the evidence for each flag.

Open →

Method

Which figures are measurements, which are estimates, which are archives.

Read the method →

5

Questions

Does Strategy Arena trade for me?

No. It is simulation and research. It never connects to a broker account.

Does “survives” mean the strategy will make money?

No. It means the strategy passed the sealed test on past data. It is a reason to keep studying it, not a forecast.

Why does the same strategy pass on one market and fail on another?

Markets do not behave alike. A rule that fits oil can fail on Bitcoin. This is why the test names the market and the bar size every time.

What is free?

The Lab with its local proofs, the MCP connection, the online judge, Quick Check and the public arena. Compare plans

Past results on historical data. Nothing here is investment advice or a forecast.