FAQ · Evaluating AI investing tools

How do you evaluate an AI investing tool?

Reviewed against the platform's code on Sep 26, 2026

Evaluating an AI investing tool means checking the evidence behind its output, not the claims around it. Six questions do most of the work: is there a record published before the outcomes, are its probabilities calibrated and is that shown, are costs modelled and stated, who holds the money, are the method and its failures published, and is it advice or research?

Why it matters

An AI model will find patterns in almost any market history, including patterns that are pure chance, and a tool built on one is often judged by results chosen after the fact: a backtest tuned until it looks good, or the best of many forecasts. The more variations were tried before one was shown, the more of a strong backtest is luck (Bailey, Borwein, López de Prado and Zhu, 2014). The six questions ask for evidence that cannot be arranged once the answer is known. A record written before outcomes cannot be edited to fit them; a stated probability can be checked against how often events happened; costs decide whether an edge on paper survives trading. Custody and advice matter for another reason: they decide who can move your money and who answers for a decision.

How it works

Ask six things. First, is there an out-of-sample record, published at the time and before the outcomes, that keeps the periods that went badly? A backtest, however long, is not one. Second, are the forecasts calibrated: when the tool says 70%, does the event happen about 70% of the time, and does it show that comparison, for example with a reliability table or a Brier score (Brier, 1950; Gneiting, Balabdaoui and Raftery, 2007)? Third, are trading costs modelled and stated: spread, commission, market impact and borrow on shorts, with every result labelled gross or net of them? Fourth, who holds the money: a tool that takes custody or trades your account carries different risks from research software. Fifth, is the method documented and are failures published, so a reader can see what was tried and what did not work? Sixth, is it advice or research: who decides, and who answers for the decision?

How Opulence Alpha applies it

Opulence Alpha's answers, from what the site already publishes. A record before outcomes: every Regime Radar forecast is stored the day it is published, never edited, and resolved five sessions later in Prediction Records, and every closed Golden Ticket week since January 2, 2026 is published in full on the ledger, whether it rose or fell. Calibration: once enough records have resolved, Prediction Records shows the average forecast beside the share that changed, with the Brier score. Costs: the Golden Tickets are gross of costs and published as such; the three Books are net of modelled costs, with the spread estimated from daily prices; the studies are before costs. Custody: none; the platform holds no money, connects to no broker and places no orders. Method and failures: the models are described on the Technology page and the studies' method and limits on their methodology page; none of the 24 popular trading beliefs tested showed a proven edge in the direction traders believe, and all 24 are published. Advice: none; it is research, and every decision is yours.

Forecasts recorded before their outcomes →

Questions

Does Opulence Alpha publish a record before the outcomes are known?

Yes, in two places. Each Regime Radar forecast of a regime change within five sessions is stored the day it is published, never edited, and resolved against what happened in Prediction Records. Each Golden Ticket basket is picked after Monday's close and held for the week; signed-in accounts see the names that night, and the public ledger publishes every closed week since January 2, 2026, whether it rose or fell. The studies, by contrast, are historical measurements, not a live record.

Are Opulence Alpha's forecasts calibrated, and is that shown?

The Regime Radar's chance of a regime change within five sessions is calibrated by isotonic regression followed by Platt scaling and shown with a 95% interval. Once enough records have resolved, Prediction Records shows the average forecast beside the share of records that changed, with the Brier score. The stock forecasts' 90% intervals are checked for coverage, and each day is stamped ok, degraded or uncalibrated by what was measured, not assumed.

Does Opulence Alpha model trading costs, and say which figures include them?

Yes. The Golden Tickets are gross of costs and published as such, because they measure the forecasts alone. The Growth, Moderate and Conservative Books are net of modelled costs on every fill: a bid-ask spread estimated from each stock's daily prices and capped at 60 basis points, a 0.5-basis-point commission, and Almgren-Chriss market impact at default parameters, as there are no live fills yet. The studies are before costs, and their limitations say so.

Who holds the money when I use Opulence Alpha?

You do, with your own broker. Opulence Alpha is non-custodial research software: it does not hold client money or assets, does not connect to your broker and does not place orders. Its Books and test portfolios are computed by the platform and hold no client money. Acting on any of it is your decision, placed through your own broker.

Is Opulence Alpha's method documented, and are its failures published?

The models are described on the Technology page, and the studies' data, method, tests and limitations on their methodology page; the evidence snapshot behind every study figure is published as JSON. Failures are published with the rest: every closed Golden Ticket week stands on the ledger as published, and none of the 24 popular trading beliefs tested showed a proven edge in the direction traders believe. All 24 are published.

Is Opulence Alpha advice or research?

Research. Opulence Alpha provides quantitative research and software tools; nothing on the site is investment advice, a recommendation, or an offer to buy or sell any security. Its forecasts and fair values are model estimates, and its studies are measured tendencies published with their limits. As the disclosures and terms state, decision and execution authority is always yours.

References

  • Bailey, D. H., Borwein, J. M., López de Prado, M. & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society 61(5).
  • Brier, G. W. (1950). Verification of Forecasts Expressed in Terms of Probability. Monthly Weather Review, 78(1), 1–3.
  • Gneiting, T., Balabdaoui, F. & Raftery, A. E. (2007). Probabilistic Forecasts, Calibration and Sharpness. Journal of the Royal Statistical Society: Series B, 69(2), 243–268.

Educational content about research methods. Not investment advice.