Technical due diligence pack
One page to review before a call: how the research is built and validated, where the data comes from and when, how risk is limited, how accounts and data are handled, and every machine-readable endpoint. Every statement here is already public on the site and checked against the code.
1. Methodology
- Universe: 1,767 US-listed common stocks, 1,345 listed today and 422 that no longer trade, kept with their history to limit survivorship bias.
- Forecasts: gradient-boosted decision trees (CatBoost), LSTM neural networks and XGBoost quantile models whose ranges are calibrated by conformal quantile regression, combined by Bernstein Online Aggregation. No generative AI is used in the forecasts.
- Regimes: nine axes of the US market, each read by a percentile classifier and fused by a hidden semi-Markov model into one composite reading, published every trading day.
- Evidence: model skill is scored on a Model × Context lattice of roughly 4,000 cells a day, across sectors, liquidity buckets and horizons, with empirical-Bayes shrinkage so a thin cell cannot over-claim.
- Books: the Growth, Moderate and Conservative model portfolios are sized by conviction on the 5-, 21- and 63-trading-day forecasts, run every trading day and measured net of modelled trading costs. The Golden Tickets are the weekly benchmark engine, measured gross of costs.
2. Validation: the Promotion Gate
- A strategy is validated before any Book may express it: walk-forward trials and purged cross-validation with an embargo.
- The winning variant must survive a Romano-Wolf step-down at a 5% family-wise error rate across every variant the search compared.
- Bars: Deflated Sharpe Ratio at least 0.95, probability of backtest overfitting at most 0.30, Hansen SPA p-value at most 0.05. On shallower evidence the overfitting and SPA bars tighten; they never loosen.
- A check that could not run counts as a failure.
- Afterwards, the record stays public: every Golden Ticket week on the Refutation Ledger and every Regime Radar forecast in the Prediction Records, written down before its outcome.
3. Data sources and timing
- Sources: daily prices of US-listed stocks, SEC EDGAR filings, the St. Louis Fed's FRED series, FINRA short-interest data, and analyst rating and price-target actions.
- Company fundamentals are read as of their SEC filing dates, not the period they describe. FRED series enter on the day they were published; the macro layer builds 347 indicators from 78 of them.
- Prices are daily closes, adjusted for splits and, where the source records them, dividends. Coverage is US-listed common stocks on the NYSE and Nasdaq.
- Fair value and the Regime Radar are recomputed after each US trading day, the Books run every trading day, and every page states the date or the close it reflects.
4. Governance and risk limits
- Every change a Book proposes is checked against a risk budget set in one-day 95% CVaR, at most 4.5% of the Book's value, before it is taken. A block never stops a reduction.
- Drawdown circuit breakers at 10%, 15% and 21% below the Book's peak value scale the budget down, then stop new risk; stepping back down takes three sessions and a recovery below half the tier's level.
- Regime overlays can only tighten a limit. Gross exposure never exceeds 165%; one issuer is capped at 25% of the Book and one sector at 50%.
- House limits are versioned and dated; a change needs a second person's approval and a pre-registered hypothesis, and an exemption expires within 30 days. Every risk verdict can be replayed exactly.
- Non-custodial: the platform holds no client money or assets, connects to no broker and places no orders.
5. Security and privacy
- Sign-in is through an identity provider, today Google or LinkedIn; the platform holds no password of yours. An unverified email address stops the sign-in.
- A session lasts 12 hours, and the Profile screen can end any session, or all of them at once.
- The console takes no payment details, and no payment processor is connected to it.
- Security reports follow /.well-known/security.txt. What the platform holds about a person, and what it does with it, is set out on /privacy.
6. Terms and disclosures
- Opulence Alpha is research software. Nothing on the site is investment advice, a recommendation, or an offer to buy or sell any security.
- The terms of use and the full disclosures, including how each figure is measured and its limits, are published in both languages.
7. Every public JSON endpoint
Free to read, with no account. The list was last checked on September 26, 2026.
| Endpoint | What it holds | Updated |
|---|---|---|
| /api/public/regime | The Regime Radar: nine axes, the composite reading and the chance of a regime change within five sessions | Every trading day |
/api/public/market-regimes/{slug} | One market-regime page's figures, for example /api/public/market-regimes/volatility-low | Every trading day |
| /api/public/prediction-records | Every Regime Radar forecast, recorded before its outcome, with calibration | Every trading day |
| /api/public/fair-value | Where US stocks sit against fair value: the distribution, sector medians and the ten names furthest either side | Every trading day |
| /api/public/equity | The three Books and GT — Equal against the S&P 500 and the Nasdaq 100, year to date, in percent | Every trading day |
| /api/public/books | Each Book's return, volatility and drawdown, declared policy, aggregate exposure and policy versions | Every trading day |
| /api/public/market-tape | The home page's market ribbon: closes, the day's change and recent history | Every trading day |
| /api/public/golden-tickets | Every Golden Ticket week: the basket, sides, entry and exit prices and returns | Every week |
| /api/public/evidence | The studies' evidence snapshot | Each research release |
| /api/public/evidence/drops | Every frozen Evidence Drop, each with its date and link | Each research release |
/api/public/evidence/{date} | One Evidence Drop exactly as it shipped, never changed | Each research release |
Also public: the Golden Tickets RSS feed at /feed/golden-tickets.xml, and /llms.txt and /llms-full.txt for answer engines.
Next step
A technical due diligence call walks through any of the above against the code.
Opulence Alpha is non-custodial research software. Nothing here is investment advice.