Markets with AI · Materials prepared with support from AI · Updated 3 October 2026 · Method and limits.
Assistant, analysis and performance system
A conversation that interprets a document isn't the same as a program connected to a negotiating platform. To review a proposal, identify what each party does: get data, generate a hypothesis, check rules or send an order. Don't pretend that there's a function because the announcement says "artificial intelligence."
One system can use determinist rules at some stage and a generative model at another. He wants to know where the IA interacts and what controls its exits. If the supplier does not explain its scope, that information remains outstanding. That guide teaches how to audition statements and markets with IA does not offer an connected bot or automatic run here.
A tab to check the evidence
| Question | Evidence to be requested |
|---|---|
| What product do you use? | Instrument, market, currency and specifications. |
| Where did the data come from? | Source, dates and processing of delays or absence of data. |
| What rules do you decide or run? | Observing conditions, permits and system boundaries. |
| What evidence does the result support? | Period, operations, costs, falls and selection method. |
| What happens when a component fails? | Documented conduct and operation limits. |
| What haven't been tested? | The difference between an example, a backtest and a later registration. |
How to audition a promise with IA
We'll use an invented words: "Our bot gets 90% and works with every asset." No period, number of operations, costs or definition of success. The task is to separate statements and absent evidence, not to declare that the product works or is a fraud.
Propt to check a description
That's what I'm talking about. That's what I mean. That page doesn't check a model automatically.
Why an isolated success rate isn't good enough
An arithmetic example: nine gain of 1 €and a loss of 20 €produce − 11 €in ten operations, even before adding costs if these amounts were gross. There are 90% success and negative result. The proportion of winners does not replace gain size, losses, costs and exposure.
He's always asking for the definitions. An open operation does not amount to an actual result. A select benefit from a favourable period does not necessarily describe the complete registration. The IA should point to these differences and demand evidence, not choose the most attractive interpretation.
Exercise: 90% of success shows profitability?
No. In a fictitious case, 9 × 1 − 20 = − 11 €. Also we have to check how they have been defined as good, cost and period.
Adaptation to different products
That a program accept a symbol doesn't prove that it models its units or its performance correctly. For forex you require the pair and currency of the result, for futures, contract size and movement value, for crypt, market and concrete product, and for actions, quantity and information corresponding to the instrument.
The comparison should preserve these differences and check the documented system boundaries. If there's only one demonstration with an asset, don't make them evidence from everyone else. ESMA's warning about IA and investments explain why information and mistrust of performance promise should be contrasted.
Practice the complete process
Solves tasks with data, prompts and independent controls at free course.
Go to related courseContinue with another task
Sources to check
The examples are computed with fictitious data. The sources do not support a sign or a performance result of these exercises.