An algorithm needs more than a story.

The adjective does not explain the investment

“AI-powered,” “institutional” and “proprietary” may describe aspects of a product, but none tells you what makes it buy, sell or remain in cash. If the most detailed explanation is a technology label, the main investment questions are still unanswered.

The CFTC’s advisory on AI trading bots warns against claims of extraordinary or guaranteed returns and points out that AI cannot predict every future market change. That is a reason to investigate the process behind a claim, not to assume that every use of machine learning is illegitimate. Read the CFTC advisory.

The better starting point is a sentence about the strategy itself. What recurring behaviour or economic relationship does it try to use? What instruments express that idea? What conditions could make the relationship stop working? Those questions make an attractive story testable.

Explainability has several layers

The first layer is the investment thesis. A momentum strategy and a macro allocation model can hold some of the same securities for very different reasons. Understanding the reason helps you interpret their behaviour when the market changes.

The second layer is construction: inputs, selection rules, weighting, rebalancing and constraints. The third is implementation: decision times, supported instruments, order handling and account requirements. The fourth is evidence: what was tested, on which data, at what costs, and whether the record represents a simulation or actual account activity.

You may not receive every parameter or the underlying source code. Proprietary details can remain confidential. The useful question is whether the available explanation is sufficient to evaluate the risks you would take. When a material assumption is unavailable, treat it as an unresolved research question rather than filling the gap with a favourable guess.

A polished backtest can answer the wrong question

Imagine testing hundreds of small variations and publishing only the strongest historical curve. The winning version might capture a durable relationship, or it might be especially well matched to that sample. The final chart alone cannot tell you how many unsuccessful variations preceded it.

Ask what was chosen before the test and what changed after the results were seen. Ask whether a later period, a different market environment or plausible changes to costs were examined. An assumption that produces a fragile result deserves attention even when the headline percentage is impressive.

This is a research framework, not a claim about a named provider’s methodology. Nor does a good research process guarantee a good future return. It helps identify how much confidence the historical evidence can reasonably support.

Versions matter when the market gets difficult

A strategy described in one document can differ from the version now running. Inputs may change, a data source may be replaced, or an execution constraint may be introduced. If you compare a current explanation with an older performance chart, you need to know whether they describe the same product.

Keep the document you reviewed and note its version or date. When a change is announced, ask which part of the process changed and whether the historical evidence has been recalculated. A revised backtest and an unchanged live record should remain distinguishable.

A losing period is also a test of communication. A useful explanation identifies whether the process behaved as designed and what assumptions were challenged. A promise that the next version will make everything back is not a substitute for that analysis.

Where the decisions sit in SquaredQ

SquaredQ’s documented strategies have centrally defined security selection, raw weights, entry and exit logic, and timing. Customers choose from standardized products and set their own allocations and account-level risk controls. The Reader then uses the account’s NAV, holdings, prices and broker constraints to calculate quantities.

This distinction matters. Choosing a product is not the same as approving each underlying security trade. After affirmative activation, the configured software can place orders without a separate per-order click. The customer’s allocation and risk settings influence execution; they do not turn the product into a personally designed investment recommendation.

The Data Room provides versioned factsheets and supporting product documents. Portfolio Studio keeps hypothetical exploration separate from live configuration. Together, those parts of the platform provide a structured way to study the process before enabling it.

How SquaredQ addresses the black-box question

SquaredQ’s answer is to organise the explanation around the actual strategy, its research and the account workflow. A product name leads to a thesis and construction guide; the guide leads to current documents; the chosen product leads to explicit execution settings in MyAlgo.

That is a practical route to understanding, not a promise of complete source-code disclosure, independent verification or profitable results. Start with a factsheet. If you cannot explain what the strategy is trying to do and the circumstances in which it could lose, use that as the next question for the team before activating.