Deterministic vs. AI: What "Reproducible" Actually Means

A lot of trading tools now describe themselves as "AI-powered." The AIQ Expert Rating is deliberately built the opposite way — and the difference matters more than it first appears, especially if you ever plan to test or trust the output.

No language model runs when your rating is computed

When the indicator scores a stock, there is no large language model involved at that moment. The rating is produced by a fixed set of rules evaluating price, volume, money flow, and trend — the same rules, in the same order, every time. Given the same input data, the engine returns the same number today, tomorrow, and a year from now.

That property has a name: determinism. Same input, same output, always.

Why that's the whole point

Determinism is what makes a rating reproducible — and reproducibility is what makes it testable. If you want to know whether a rating meant anything historically, you can run it across years of data and get one, stable answer. The rule that fired on a bar in 2019 fires the same way if you re-run it today. Nothing drifts.

Compare that to a tool that asks a language model to look at a chart and form an opinion. Ask it the same question twice and you can get two different answers, because that kind of model samples from probabilities rather than following fixed rules. It may be articulate. It may even be right. But you cannot backtest an opinion that changes each time you ask for it, and you cannot audit a decision the tool can't reproduce.

"AI at build time, math at runtime"

None of this means AI has no place. The right division of labor is to use modern tools to help design and explain a rule set — and then to freeze that rule set so that, at the moment a rating is computed, only deterministic math runs. Build with whatever helps. Decide with rules you can reproduce.

That's the tradeoff AIQ makes on purpose. You give up the loose, conversational flexibility of a model forming fresh opinions. In exchange you get a rating you can test against history, reproduce on demand, and trace back to the exact evidence that produced it.

What this buys you as a trader

Three concrete things:

• You can backtest it. A reproducible rating can be evaluated honestly over historical data, because it doesn't change between runs.
• You can audit it. Every rating traces back to the specific conditions that fired to produce it — no black box, no "the model felt bullish."
• You can trust the number tomorrow. The rating you see isn't a fresh guess each session. It's the same rule set that has always run, applied to today's data.

In a field full of tools that generate confident-sounding opinions, a rating you can reproduce and check is a higher bar — not a lower one.

Deterministic. Auditable. Reproducible. Explainable AI, before the term existed.

← Back to all research