Android prototypes: Pocket Lab & Time Machine · Experimental offline APKs, separate from the Windows Lab.

Lab 253: the judge · Windows · Linux · Free

Strategy Arena LabFree Pine Script strategy tester and backtester, for Windows and Linux

Strategy Arena Lab is a free app for Windows and Linux that tests a TradingView (Pine Script) strategy out of sample: the end of the history stays sealed during the search, every variant tried is counted, trades are compared one by one with TradingView's, then the judge returns its verdict: holds, mirage or inconclusive.

Your AI drives the engine, your GPU searches, the judge decides. Same trades as TradingView, then data nobody has seen: winner or mirage? Free local testing, no live orders.

Windows 10/11 x64 · Linux
Data stays on your machine
NVIDIA GPU optional

The Lab installs on a Windows or Linux PC

You are on a phone: copy the link of this page to open it on your computer. Meanwhile, Prisme, our free Bitcoin chart workshop, opens right here in the browser, no account.

Open Prisme online

Download free — Windows EXE MSI package Watch the demo (2 min)

Windows shows “Windows protected your PC”?

The installer is not signed yet, so SmartScreen warns about any unsigned program. Click More info, then Run anyway. Before that, you can check that the file is the published one: in PowerShell, Get-FileHash on the downloaded file must give the SHA-256 shown in Verify before installing.

Linux: version 253 (CPU engine, no GPU): AppImage · .deb package — start-up checked on Ubuntu; the full interface was not tested on Linux.

Version 0.1.0-253 · build of 27/10/2026

New in Lab 250 (and 243):

Ideas from Prisme can be judged: import a Prisme card (level, zone, saved filter, double top or bottom), the Lab turns it into a testable rule; periods already seen in Prisme are logged and cap a certified verdict.
When the Pine you exported was checked in TradingView, the signed certificate says so (sealed data and receipt), ready for a public verification page.
Compact answers for your AI stay under 12 kB by construction, and French labels are completed.
A fairer judge: compared with buy-and-hold over the test period itself, trials counted as effectively independent, and any return series can be judged — your own bot, a portfolio or an external strategy.
Many more strategies can be written: daily filter on a 4h script, arithmetic between indicators, risk-based sizing, break-even and ATR trailing exits.
A new strategy idea can again earn “survives”: the sealed test period is counted per rule family, and stop and target distances are now searched by the GPU like any parameter.
Short margin calls on gap bars match the real TradingView (45 of 45 trades), and the Windows program is a third smaller.
TradingView rules by default: the Pine you export is exactly the strategy the Lab judged. Checked in the real TradingView: 234 of 234 trades on the leaderboard, 263 of 263 with Bollinger and MACD, 361 of 361 with short margin calls.
More reliable statistics: computed on the account equity the engine really held (position size, fees, open positions), stricter verdicts, proofs written atomically and checked on read.
Weekly challenge, “My idea in plain words” without AI, portfolio of 2 to 6 strategies, and any market added from public Binance or Crypto.com candles.
“Winner or mirage?”: a sealed proof on a period unseen during the search, against Buy & Hold on the same windows, counting every variant tried.
Any AI at the controls: an essential MCP profile, compact answers and start_here, so that small local models complete the path too.
Background GPU jobs: large sweeps run on your NVIDIA card while the CPU reference replays the rows that are kept.

Local research and JSON export stay free. A saved local report is not an independent paid audit.

Plug your AI into the Lab (Claude Code, Codex, opencode with DeepSeek, Cursor, LM Studio): setup, the path to a judged Pine, and what the AI cannot do

How the proof is kept rigorous

Pine is an import bridge for deterministic subsets; Strategy Arena generates and measures ArenaScript first. Rust CPU is the final verification authority, CUDA only accelerates search, and INCONCLUSIVE remains a valid scientific outcome.

Official public beta channel. Local quick proofs and JSON export are free. No broker connection and no live orders. Integrity is by published SHA-256 (this channel may be unsigned until the Windows signing gate clears).

One-time install code

Paste this code in Strategy Arena Lab under Account and Install attribution. It expires after 30 minutes.

Preparing code...

Pine Script

A free Pine Script strategy tester that runs on your PC

TradingView’s Strategy Tester shows the result of the settings you kept. It does not know how many settings you dropped, nor how the strategy behaves on data the search never saw. The Lab answers those two questions.

01 · PASTE
Your Pine strategy, imported

Paste the Pine Script of a strategy. For supported Pine, the Lab rebuilds the same rules and its trades are compared one by one with TradingView’s list of trades.
02 · BACKTEST
Real market history, fees included

Backtest on the market history shipped with the Lab or on any market added from public Binance or Crypto.com candles, against Buy & Hold on the same windows. Every variant you try is counted.
03 · JUDGE
A verdict on data never seen

The end of the history stays sealed during the search. The judge then rules: holds, mirage or inconclusive. You can export the judged strategy back to Pine Script.

Free: local research and JSON export
Windows 10/11 x64 (EXE, MSI) · Linux (AppImage, .deb, CPU engine)
Runs on your PC; NVIDIA GPU optional for large sweeps
No broker connection, no live orders

Research on past data, not investment advice. An unfavorable verdict is still a result. Why a Pine strategy wins in backtest and loses live

Plug your AI into the Lab in 3 steps

The Lab speaks MCP, the protocol that lets an AI use software. Your AI client starts a small local script; the script talks to the Lab running on your PC.

Start the Lab first. It opens a local connection on 127.0.0.1 with a new token for each session.
In the Lab: More tools → AI control → Copy the MCP config. The Lab fills in the real path of the script on your PC.
Paste it in your AI client (examples below), restart the client, then ask: “call start_here”.

Requirement: Node.js 20.3 or newer for the connection script. Replace PATH_COPIED_FROM_THE_LAB with the path the Lab gives you.

Claude Desktop

File: %APPDATA%\Claude\claude_desktop_config.json (Windows).

{
  "mcpServers": {
    "strategyarena-lab": {
      "command": "node",
      "args": ["PATH_COPIED_FROM_THE_LAB"]
    }
  }
}

Cursor

File: ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project).

{
  "mcpServers": {
    "strategyarena-lab": {
      "command": "node",
      "args": ["PATH_COPIED_FROM_THE_LAB"]
    }
  }
}

LM Studio

In LM Studio: Program → Install → Edit mcp.json. Pick a model that supports tool calls.

{
  "mcpServers": {
    "strategyarena-lab": {
      "command": "node",
      "args": ["PATH_COPIED_FROM_THE_LAB"]
    }
  }
}

opencode

File: opencode.json in your project (or ~/.config/opencode/opencode.json).

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "strategyarena-lab": {
      "type": "local",
      "command": ["node", "PATH_COPIED_FROM_THE_LAB"],
      "enabled": true
    }
  }
}

Claude Code

claude mcp add strategyarena-lab -- node "PATH_COPIED_FROM_THE_LAB"

What your AI sees: the essential profile

By default the Lab shows a short list of tools with compact answers, so that small models can drive it too. start_here gives the whole path with ready-to-send arguments.

start_here
list_sandbox_datasets
validate_sandbox_strategy
run_sandbox_backtest
run_sandbox_sweep
start_sandbox_job
sandbox_job_status
export_sandbox_pine
analyze_pine_strategy
compare_tradingview_trades
prove_sandbox_script
begin_local_quick_proof
list_sandbox_experiments
suggest_lab_capability

Need every tool? Add "--profile", "full" after the script path in args (or set STRATEGY_ARENA_MCP_PROFILE=full), then restart the client.

First session

Build an idea. Compare it. Test it on unseen data.

A train result is not KEEP. Silver Buy & Hold can print near +99% on the frozen tape — you have to see it.

01 · ASSEMBLE
Ride SMA20, dual SMA, Event OR/AND or Turtle

Ride SMA20 and dual-SMA oil let winners run without a take-profit. Event Proof keeps its OR seed and adds an AND companion; Turtle uses a Donchian-20 breakout. The 73 public fighters stay a strategy dictionary.
02 · PLAY
See strategy versus Buy & Hold

One chart: close, strategy equity, Buy & Hold, SMA200, cash, trades, and a curtain on the sealed 20%. Three bosses: cash, the tape, SMA200. Play never opens holdout.
03 · THINK, THEN JUDGE
Search for variants, then run the final test

Think ranks neighbours that beat Buy & Hold on the full train, then checks internal folds. overfitSuspect is a success, not a failure. KEEP / REJECT / INCONCLUSIVE only after Judge. TradingView stays optional and never buys a greener verdict.

Trust

Verify before installing

This beta is unsigned. Windows SmartScreen may warn you; download only from this official page and compare the SHA256.

Release manifest Signature report

Channel: public beta
Package: 0.1.0-253
Signature: unsigned
SHA256 EXE: 0fcd7fd6e57f4182ab7fb03a0d05ad9fd9208365613ac2aea9f7666b953d1883
SHA256 MSI: af8130cd8edf95b71f309a780cbb925270914723bc9380dcdb554b7a5a15b9ba
Technical build ID: 20261027-253-n1

Launch video · 73 s

Strategy Arena Lab in 73 seconds: your AI searches, our judge decides

Demo: scripted AIs, real Lab computations

Silent video with on-screen captions. An illustrated opening (Grok Imagine), then real captures of the Lab window: in the Colosseum, Claude, Cursor and DeepSeek take one challenge while a period stays sealed; the GPU sweeps 200,000 variants in 2.8 s and the CPU replays the ranking with 0 mismatches; Night mode delivers its morning report (verdict: Mirage); the TradingView judge replays 234 / 234 identical trades; the reveal shows a training winner that is a mirage on the sealed period; a published badge is recomputed by the server on strategyarena.io/verified. Past results, not a forecast.

Inside the app

See the Lab before installing

Play the frozen tape against Buy & Hold before any KEEP. Rust CPU still judges.

Strategy Arena Lab Play chart: Ride SMA20 equity finishes above Buy and Hold on frozen silver, holdout curtain sealed — **Watch green finish above gold**Ride SMA20 on frozen silver: strategy +136.8% vs Buy & Hold +99.5%. The right curtain stays sealed. Play is not KEEP.

Three bosses panel: strategy +136.78%, Buy and Hold tape +99.49%, vs B&H +37.29% on silver train — **Read the three bosses**Cash, the tape, SMA200. Beating one boss on train is not KEEP.

Strategy Arena Lab assembler: Ride SMA20, Play Think Judge, sealed silver tape — **Assemble, then Play**One click loads Ride SMA20. Play, Think and Judge stay on one row. No broker, no live order.

Native Rust Monte Carlo diagnostic in Strategy Arena Lab — **Add uncertainty context**Monte Carlo remains a diagnostic after selection and never replaces the holdout.

TradingView Starter

Choose the account path. Prove the strategy separately.

Your first chart and price alert do not require the Lab. Strategy Arena proof remains required before trusting or exporting a strategy.

Partner attribution is for first-time users and web purchases only. Strategy Arena may earn a commission on eligible purchases; starting with a free account needs no payment.

I have never had an account · Start free I already have an account

Plans

Start free. Choose what you need next.

Keep testing on your PC for free. Choose Builder to keep reports and versions, Operator for larger supervised research, or a separate 9.90 € server audit. No audit purchase is required for a subscription.

Free · Quick Check + Lab · 0 €

Check an idea on the website or run the complete local proof. No account or card required.

Run the free Quick Check

Survival Audit · one-time · 9.90 €

Independent of your PC: sealed holdout, private URL, report you can reopen. A bad verdict is still a delivered audit.

Get the 9.90 € report

Builder · Keep 9 EUR/month

Includes Free. Return to every useful report, strategy version and reusable export.

Explore Builder

Operator · Delegate 19 EUR/month

Includes Builder. Let an AI run bounded, sequential and auditable research campaigns under your control.

Explore Operator

Payment never improves the verdict, holdout or uncertainty evidence.

Cloud Research Run, Edge Monitor, Health Score, Leagues and Portfolio Arena have separate limits. Public report publication requires server checks; a saved local report is not an independent paid audit.

See limits and subscription details

FAQ

Before your first run

Is an NVIDIA GPU required?

No. The CPU reference engine works without a compatible GPU. CUDA is used only when it is available.

Do Play or Think open the holdout?

No. Play and Think stay on the train. KEEP / REJECT / INCONCLUSIVE exist only after Judge opens the sealed 20%.

Where does my data go?

Backtests stay on your machine. Synchronization starts only after you explicitly sign in and choose to sync.

Why can SmartScreen show a warning?

The current beta may be unsigned. Use the official installer link and compare its SHA256 with the value published above.

Is this a live trading robot?

No. Strategy Arena Lab is an educational research and backtesting tool. It does not place broker orders or provide investment advice.

Educational research software. Simulated results are not investment advice and do not guarantee future performance.