Documentation

Strategy Workflow

A useful Astral strategy workflow has four phases: build, inspect, validate, and operate. Skipping inspection is the fastest way to save or deploy a strategy you do not actually understand.

1. Build the first draft

Start with a complete prompt. Include:

  • symbol or asset group,
  • timeframe,
  • entry condition,
  • exit condition,
  • risk controls,
  • whether you want alerts, backtesting, or executable order logic,
  • any constraints such as long-only, regular market hours, or avoiding overtrading.
Specific prompt

"Build a long-only SPY 15m strategy. Enter when price reclaims the 20 EMA after RSI crosses above 50 and volume is above its 20-bar average. Exit on a 2% stop, 4% take profit, or RSI below 45. Run a backtest."

Weak prompt

"Make me a profitable strategy."

This gives the system too much room to invent rules, risk assumptions, and evaluation criteria.

2. Inspect the generated logic

Before trusting a result, ask:

  • What triggers an entry?
  • What triggers an exit?
  • Is the strategy long-only or does it imply short exposure?
  • Does it use one asset, multiple assets, or a basket?
  • Does it depend on intraday intervals or daily bars?
  • Does it require indicators that are rendered on the chart?
  • Does it include stops, take profit, time exits, or rebalancing?
  • Does any rule reference unavailable data?

If the strategy is unclear, ask Astral to explain it in plain English before changing it.

3. Review the backtest

Backtest metrics should be read together.

  • Return is not useful without drawdown and trade count.
  • Win rate can be misleading if average loss is much larger than average win.
  • Sharpe/Sortino can be unstable on short samples.
  • Low trade count often means the result is not representative.
  • Very smooth equity can be a sign that the strategy is not doing what you think.

Use the chart and code output together. If a trade marker, indicator, or equity curve does not make sense, investigate before saving.

4. Save, publish, or deploy

After review, decide what the strategy is for:

  • Save if you want to keep it private and revisit it later.
  • Publish if it is useful as a shared Explore asset.
  • Deploy for alerts if you want rule-triggered notifications.
  • Deploy to broker only if your account, plan, broker connection, verification state, asset class, and risk assumptions are ready.

Deployment is not validation

Deploying a strategy means Astral will monitor or operate it according to the selected mode. It does not prove the strategy is profitable, appropriate, or compatible with every broker/account restriction.

Revision patterns

Use targeted revision prompts:

  • "Keep the entry condition, but replace the exit with a trailing stop."
  • "Reduce trade frequency by requiring trend confirmation."
  • "Show me why the last backtest changed."
  • "Add a volume filter without changing the stop."
  • "Run the same strategy on QQQ and compare the result."
  • "Convert this strategy into an alert condition."
  • "Make this long-only and regular-hours only."

Common failure modes

The strategy overfits

If one parameter value produces most of the result, broaden the test or simplify the rule.

The rule is ambiguous

If the prompt says "strong momentum" or "good trend" without defining it, the generated logic may not match your intent. Ask for explicit thresholds.

The backtest looks too good

Check data range, trade count, position sizing, exits, and whether the system accidentally allowed future-looking behavior.

The strategy is not broker-compatible

Broker deployment has restrictions around asset class, account type, long-only behavior, available buying power, fractional support, account integrity, and live account rules.