An AI trading forward test should leave you with more than a green number. You want to know which configuration ran, what decisions it made and whether you would understand the same behaviour during a losing period.
That is the practical next step after choosing Astra in Alpha Pulse AI. The integration can operate correctly while the trading idea performs poorly. A decision can also sound convincing while relying on a mistaken interpretation. Treat those as different questions.
Review Alpha Pulse AI and its setup requirements before setting up the evaluation. It connects external models to MetaTrader; it does not remove the need to examine risk, costs and the difference between a demo environment and live execution.
Give the AI trading forward test a fixed starting point
Save the EA version, selected model, prompt or strategy engine, symbol, timeframe and input settings. Record the broker and account type too. Give this combination a short configuration name so every observation can be traced to it.
If you change the prompt, risk input or model during the run, start a new version in the record. Keep the old observations attached to the old setup. Mixing them into a single total makes it harder to answer what produced the result.
You can use a simple document. You do not need to build a research platform before the first session. What matters is that the record stays complete when a trade disappoints you, not just when you want to share a screenshot.
Ask three questions in the right order
First, does the setup operate? Check the intended model, API connection, request timing and response. Confirm that the EA is attached to the correct account and chart. Investigate connection errors and unexpected settings before interpreting the trading results.
Second, does its behaviour fit your brief? Read the reasoning available in the panel. Compare the interpretation with your written criteria. Record both decisions to trade and decisions to stay out. If the model never trades, that is something to investigate, not a reason to lower every threshold immediately.
Third, what economic evidence has accumulated? Look at trading costs, open equity, closed results and the operating cost of the model. The first two questions help you interpret this evidence; neither answers it for you.
Keep a small record you will actually maintain
| Record | What it helps you check |
|---|---|
| Time and configuration version | Which settings were active when the decision happened. |
| Decision and available explanation | Whether the interpretation matches the method you intended. |
| Order, size and execution | What the EA actually placed, rather than what you expected. |
| Open equity and closed result | Whether unrealised losses are hidden by a balance-only view. |
| API usage and trading costs | Whether the complete workflow makes economic sense. |
| Intervention or revision | Whether your changes influenced the observations. |
Use screenshots when they preserve an explanation you cannot export conveniently. Capture the difficult examples too. A folder containing only attractive entries is a selection of outcomes, not a record of the run.
If a capital programme is part of your longer-term plan, research Axi Select separately. Its requirements belong in another part of your research; they are not validated by a successful API connection or a profitable demo session.
Affiliate disclosure: I may receive a commission if you sign up through the Axi links in this article.
Read costs and exposure together
Separate broker trading results from model charges before combining them. Use one reporting currency and state how you converted any API bill. A small trading gain can look different once its operating costs are included.
Also inspect the worst open-equity point you observed, simultaneous positions and any concentration in one market direction. A clean list of closed trades can miss a difficult period while several positions were open.
These are historical observations from that configuration. They do not establish a ceiling on future losses. Stops, account limits and demo fills should not be treated as guarantees about execution through gaps or disruption.
A short demo can answer the wrong question very well
A week may tell you that the connection works, that the panel is readable and that the setup can place orders. It may contain too few trades or too little variation in market conditions to support a useful performance conclusion.
Write separate conclusions. “The integration operated as configured” can be supported while “this method has a durable edge” remains unresolved. A quiet period also belongs in the record. Do not manufacture a sample by changing the rules whenever activity slows.
The MQL5 WebRequest documentation states that the function is unavailable in Strategy Tester. A historical chart exercise can help develop your criteria, but it does not replay the live Astra API workflow.
Decide what would make you investigate
Before starting, write down operational reasons to pause: the wrong account, missing risk controls, repeated request failures or orders that do not match the intended configuration. Define your review process for trading losses separately, with an explicit risk budget.
If you later consider external capital, compare that process with the Axi Select programme overview and the provider’s current conditions. Confirm permitted systems directly. A programme’s marketing page cannot substitute for the rules that apply to your account.
At the review date, keep the conclusions modest and useful: what operated, what disagreed with your intention, what it cost and what still needs observation. That document is an asset you can build on even when the period was unprofitable.
Why live behaviour deserves its own review
This earlier video explains why a convincing demonstration and live behaviour can diverge. Treat it as context for your review process, rather than evidence about the Astra configuration you are testing: Why Most Trading Bots Fail Live (And What Adaptive AI Does Better).
Buy the workflow only if you will use it
Alpha Pulse is useful when you want a working bridge between AI decisions and MetaTrader and are willing to supervise a documented evaluation. If you need a promise about next month’s profits, this exercise cannot supply one. What it can supply is a clearer account of what you tested and what remains unknown.
Review Alpha Pulse AI and its setup requirements
A future capital-allocation application needs its own evidence and rules review. Read the Axi Select overview if that route interests you. A demo test, including a positive one, does not establish eligibility or permission to use third-party EAs.
For more examples of the questions to ask before trusting a trading configuration, Join the DoItTrading newsletter.
Frequently asked questions
Can MetaTrader Strategy Tester validate the live Astra API workflow?
No. MQL5 documents that WebRequest is unavailable in Strategy Tester. The live external-model workflow needs a forward evaluation.
How many demo days prove that the strategy works?
There is no fixed number of days that proves it. A short run can establish that the setup operates; performance assessment also depends on sample size, market conditions and the changes made during testing.
Should I include AI API charges in the evaluation?
Yes. Record model usage separately from broker trading results and review the total operating cost. Keep the evaluation’s currency conversions consistent.
