backtrade.dev

Your strategies

Create a strategy or load a Python file to get started.

Strategy editor & replay setup

1. Save a strategy version

Python · standard library

StrategyStrategy name

strategy.py

"""Illustrative long-only strategy; not a claim of expected profitability."""
class Strategy:
    requirements = {"order_types": ["market"], "time_in_force": ["day"], "asset_classes": ["us_equity"]}
    def on_start(self, context):
        self.pending = False
        context.log("Moving-average strategy started")

def on_bar(self, context, bar):
        period = int(context.parameters.get("period", 20))
        quantity = float(context.parameters.get("quantity", 5))
        history = context.history(period)
        if len(history) < period or self.pending:
            return
        average = sum(item["c"] for item in history) / period
        held = context.position()
        if bar["c"] > average and held == 0:
            context.submit_order(side="buy", qty=quantity)
            self.pending = True
        elif bar["c"] < average and held > 0:
            context.submit_order(side="sell", qty=held)
            self.pending = True

def on_fill(self, context, fill):
        self.pending = False
        context.log("Confirmed fill: " + str(fill["side"]) + " " + str(fill["filled_qty"]))

def on_finish(self, context):
        context.log("Decision window ended; remaining holdings will be marked to session end")

Load .py file

Callbacks: on_start, on_bar, on_fill, on_finish. Continuous runs also offer on_day_start and read-only on_day_end. The runner supplies available history and manages time. Source is private to your account; keep credentials out of it.

2. Configure a replay

Independent capital per date

Holding period

Starting capital · USDCost · bps per fill

Experiment nameSaved version

SymbolExecution data

Decision barsAvailable date

Decision window

Uses each date’s market open and close, including early closes and daylight saving.

Parameters · JSON

Research setup · optional

Record how this code was written. These declarations are saved with the experiment; the runner executes Python, not these prompts. Keep credentials out.

Record prompts and model

Model decisions · optional

Call a model from Python and record its decisions. Ordinary replay needs no provider account.

Model service is unavailable. Regular code strategies remain available.

Enable model decisions

Provider settings

Your key is sealed in this browser before storage. Prompts go to Anthropic only after a separate run approval.

Connection nameAnthropic API key

Enrollment signing fingerprint

Unavailable

Compare with the deployment documentation. Browser sealing protects against passive exposure; it does not protect a compromised browser application.

Risk and runtime limits

Maximum position · sharesMaximum order · sharesMaximum loss · %

Wall-time budget · seconds

Runs execute on your computer. Queue the experiment, then start the local runner.

Starts with $100,000 per date. Extra cost: 1 bps per side. After the decision window, replay continues to the session end for scoring. Remaining holdings stay open.

Local runner setup · Docker + Node

Run on your computer

Docker + Node 22 or later

The API queues work. Your local runner executes Python in a network-isolated container without provider or broker credentials. Optional model decisions use a controlled, recorded connection. It processes only the experiment you select.

Download runner ↓Checksum ↓Installation & account key ↗

sha256sum -c backtrade-runner.mjs.sha256
node backtrade-runner.mjs --prepare
node backtrade-runner.mjs --experiment <experiment-id>

Experiments

0 recent · select up to 4 to compare

Save a version and queue your first replay. Results will appear here.