Five intelligences.Every one of them has to prove it.
Opulence Alpha is a quantitative research platform built the way an institutional research desk is governed. One service forms a view. A second decides what risk that view may take. A third searches for better policy. A fourth measures, afterwards, whether any of it was real. No service may act on its own conclusion, and no service may publish a number it does not also validate.
The result is research you can interrogate rather than a recommendation you have to trust.
These are not returns. They are measurements of whether the platform's stated confidence can be believed. A system that says 70% and is right 70% of the time is usable. A system that says 70% and is right 50% of the time is not, however good its headline looks.
How these are measured →The hard part was never the prediction.
Any competent model can produce a number. The difficulty in quantitative investing has never been generating a forecast — it has been knowing which forecasts to believe, how much size they justify, and whether last quarter's edge was skill or the residue of a thousand backtests.
This is why institutional research desks are organised the way they are. The analyst who forms a view does not set the position limit. The risk officer who sets the limit does not choose what is worth holding. The person who decides, months later, whether the view was any good is independent of both. The separation is not bureaucracy. It is what stops a confident wrong answer from becoming a large wrong position.
Software has mostly not been built this way. A single model produces an output, the same system sizes it, and the same system reports on how well it did. Nothing in that loop can catch the model being confidently wrong, because nothing in it is adversarial to the model.
Separation of powers, implemented in software.
Opulence Alpha runs as five independent intelligence services. Each owns exactly one question. None can answer another's. One rule binds all of them: a service may publish a quantity only if it also validates that quantity.
Risk Intelligence is the clearest illustration. It is the only service on the platform whose objective function is not return. Its entire purpose is to answer whether a proposed change is permissible and at what size, and it is structurally incapable of preferring a position because that position looks profitable. When the risk budget binds, it decides where the cut lands — and nothing in the belief layer can overrule it.
A service may publish a quantity only if it also validates it.
That single rule settles every boundary argument in the system. Risk Intelligence backtests its own VaR — Kupiec, Christoffersen, Basel — so it owns the covariance matrix, the factor loadings and the tail parameters. Nobody backtests crowding as a risk quantity, so Risk Intelligence does not own crowding.
A model cannot reach production by being persuasive.
Most quantitative failure is not modelling failure. It is the failure to distinguish a real edge from a pattern found by looking hard enough at historical data. A researcher who tries a thousand strategies will find several that look excellent and are worth nothing.
Opulence handles this in the promotion pipeline rather than in review. A candidate model must clear three independent statistical hurdles before it can serve a single prediction, and the pipeline is what enforces them.
Edges that survive only because many were tried
Sharpe ratios manufactured by selection
Strategies tuned to their own test window
Beneath the gates, 7,050 hyperparameter trials run each week under walk-forward validation, and training data is partitioned by market era so that models never see instruments that did not exist at the time. A model trained on 1987 has no VIX, because in 1987 there was no VIX. Look-ahead bias is excluded at the data layer, not corrected for afterwards.
A model that fails these tests does not get a second look from an analyst. It simply cannot be promoted.
Every forecast arrives with the range it could be wrong by.
A point estimate is close to useless for allocating capital. “This position returns 4% over 21 days” tells you nothing about whether to hold 1% or 8% of your book in it. What matters is the dispersion around that number and whether the stated dispersion can be trusted.
Opulence publishes beliefs as distributions. Every forecast carries a prediction interval built with split conformal prediction under adaptive conformal inference — a method that provides distribution-free coverage guarantees in finite samples, without assuming the market is well-behaved.
Financial series are not exchangeable. Volatility clusters, regimes break, structure shifts. Adaptive conformal inference handles this by adjusting its thresholds as conditions change, with separate calibration sets per regime. Intervals widen when the market is genuinely less predictable and tighten when it is not. That widening is the system working, not degrading.
When the platform is uncertain, it says so, in a number you can size against.
The platform researches. You decide.
Opulence Alpha is a research system. It does not hold your assets, connect to your broker, or place orders. Nothing moves in your accounts because of anything on this platform. Every output is research you review and act on, or do not.
Publishes beliefs, valuations, regime state and evidence across the universe
Form your own view, use the platform to check it
Analyses the holdings you enter, flags concentration, factor and regime exposure, and models what a proposed change would do
Decide which changes, if any, are worth making
Re-runs that analysis every trading day and tells you when something material changes
Decide whether to act, every time
There is no mode in which the platform acts for you. That is a design decision, not a limitation we intend to remove.
Who this is built for.
Opulence is a research instrument. It rewards people who want to interrogate a view and frustrates people who want to be told what to buy.
- You already have a process and want better inputs to it
- You read a prediction interval before you read the point estimate
- You would rather see the evidence against a position than a stronger case for it
- You are comfortable being the one who decides
- You want someone to manage your money — we are not a manager and cannot become one for you
- You want a list of tickers to buy without reading why
- You are looking for intraday or day-trading signals — the horizons here run 1 to 63 days, built for swing and position work
- You need certainty; this platform is built to quantify uncertainty, not remove it