Oneiros Vision Capital Preservation Engine

The Methodology & Architecture Whitepaper

1. Executive Summary & The Philosophy of Uncompromising Capital Preservation

In the modern financial landscape, the old paradigms of wealth management have repeatedly proven inadequate. For decades, the industry relied on the assumption that a standard 60/40 allocation of equities and bonds provides sufficient safety and diversification. This foundational belief was catastrophically shattered in 2022, an era defined by a "Correlation 1.0" shock where traditional safe-haven assets collapsed alongside risk assets. Stocks went down, bonds went down, and nowhere was safe.

Oneiros Vision was built in direct response to these structural market fractures. It measures, every session, which regime the market is in and what that regime means for the risk in a portfolio, so that a portfolio manager decides with that measurement in front of them. It does not decide for them: it holds no client money, places no orders and recommends nothing.

This whitepaper is the guide to the methodology and architecture of the Oneiros Vision engine and the risk terminal built on it. By turning raw macroeconomic telemetry into one dated reading of the market's regime, with the arithmetic behind it shown, it bridges the gap between opaque quantitative algorithms and practical, daily wealth management.

Instead of measuring risk through the limited, lagging lens of standard deviation or historical drawdowns on a static portfolio, Oneiros Vision conceptualizes risk as "Economic Weather." It is a reading of the conditions today, measured rather than opined. It does not forecast tomorrow; it states, with its arithmetic, what kind of market a portfolio is standing in now.

2. Core Investment Philosophy: A Regime & Risk-Allocation Overlay

Oneiros Vision is not a security-selection system and does not tell an investment committee which individual shares, funds, or ETFs it must own. Its role is to translate the Master Danger Index (MDI) into a portfolio-level reading: which regime the market is in, and how exposed the book is to it.

The terminal classifies every holding of a client's book into one of six sleeves, the ones the regime engine reasons in. The names below are the ones on the Regime Exposure panel and in the API's holdings_classified:

The MDI supplies the regime and the rule by which one regime gives way to another. The client, RIA, or CIO decides what, if anything, to change in a book, selects every security, and remains responsible for mandate suitability, tax, liquidity, and concentration constraints. A reference ETF portfolio is used only to make the historical mechanics observable and repeatable; it is not a recommended portfolio.

What Each Regime Has Measured

The regime is a reading of today, not an instruction. What each one has been followed by, for the S&P 500 over the 21 sessions a position taken on the call holds (Performance Tear Sheet, section 2; US dollars):

The reference strategy in the tear sheet holds its risk sleeve at 100%, 95%, 20% and 0% in these four regimes. That ladder exists to make the history observable and repeatable; it is not a recommendation for any book.

Empirical Validation: The Signal Is Robust to Implementation

The claim that Option A is illustrative rather than prescriptive is testable, and has been tested. Holding the MDI regime signal fixed and varying only the risk-sleeve exposure per regime across seven materially different allocation ladders — from the shipped 100/95/20/0% (CALM/CHOP/WARNING/CRITICAL) chassis down to a muted 85/75/40/20% RIA-scale ladder — every ladder was scored against a static portfolio holding that same ladder's own average weights, isolating the signal's contribution from the sleeve composition itself.

Two findings held across every ladder tested: the drawdown-reduction contribution is consistently positive and material (+7.29 to +12.37 percentage points of maximum drawdown versus the matched static control), and the return contribution is modest (+0.86 to +1.51 percentage points of CAGR). Muting the overlay to RIA-scale tilts retained 102% of the drawdown benefit measured on the full-intensity reference chassis. The MDI's value is primarily capital preservation, and it does not depend on the specific ETF weights used to demonstrate it.

Illiquid Private Asset Integration (Proxy Mapping)

The Mechanism: For portfolios containing opaque, illiquid assets (Private Equity, Direct Real Estate, Fine Art), the system utilizes a Proxy Mapping framework combined with an Illiquidity Beta scalar. The asset is assigned a liquid, public-market equivalent (e.g., XLRE for Real Estate), and its volatility profile is scaled by an Illiquidity Beta the adviser states. The beta is 1.0 by default, the proxy's full volatility; a lower figure is the adviser's own assumption about appraisal smoothing, and nothing in the terminal validates it.

The Utility: By proxying illiquid assets and scaling their volatility, the terminal brings these large, immobile capital blocks into the book's measured risk. The book's measured exposure to the regime then includes the risk its illiquid holdings carry, which a model reading only the liquid part would leave out.

3. The Performance Record: Sixteen Years of Backtest and the Live Log

The backtest is a granular, year-by-year reconstruction of the engine's behavior, applying the strict T+1 Execution Standard. T+1 forces the sequence of Day T (Detection at market close) and Day T+1 (Confirmation and Implementation at the following day's close). This eliminates "Look-Ahead Bias", and every trade is charged a cost: 2 bps per leg in the headline figures, with the wider range in the Transaction-Cost Sensitivity study. The backtest implements this sequence exactly, and reads the two inputs that publish a day late — net liquidity and the high-yield spread — on the day they are published.

The logic was tested over a 16-year backtest (2010–2026), explicitly modeling institutional-grade round-trip transaction costs and trading every signal at the close of the following session. The system transitioned to live, out-of-sample production on July 20, 2026, and every regime call logged live since then reproduces from the current code (52 of 52 sessions to September 30, 2026). Complete performance data is maintained in the Performance Tear Sheet.

4. The Master Regime & Systematic Anchor

At its core, the engine distills sixteen US market and macroeconomic series, read as seventeen inputs, into a single structural state of the market. In the reference strategy, that state sets how much of the portfolio is held in risk assets.

The Systematic Anchor (Master Danger Index – MDI)

The Master Danger Index (MDI) is the quantitative heartbeat of Oneiros Vision. It is an absolute 0 to 100 score of systemic stress in the US market and economy, read every session. The model tracks the distance between the current MDI score and the regime thresholds (e.g., crossing 55.0 to enter a WARNING state), and the 1-Day and 5-Day Deltas show which way the score is moving and how fast. The terminal shows both, and makes no call on what happens next.

5. Pillar Telemetry: Sub-Component Diagnostics

The MDI is a weighted culmination of three underlying, heavily researched intelligence pillars. Speedometer and Sentiment carry the majority of the composite and are weighted evenly against each other; Macro Health is intentionally the lighter, slower-moving structural check. The weights are fixed in the frozen model; they are not re-fitted as new data arrives.

[SPEEDOMETER] (Primary Weighting)

The Concept: Utilizes a Cumulative Distribution Function (CDF) of volatility Z-scores (normalized over a rolling window) across Equities, Bonds (the MOVE index), Credit (High Yield spreads), and Foreign Exchange. It identifies the pure velocity of risk expansion.

The Utility: We are looking for mechanical friction and funding stress. A sharp rise means volatility is expanding across asset classes at once, the pattern of a liquidity squeeze. Because it reads bonds, credit and currencies as well as equities, it registers cross-asset stress that an equity-only measure would not show.

[SENTIMENT] (Primary Weighting)

The Concept: A Weight-of-Evidence (WoE) pillar measuring stress in positioning and hedging demand. Rising volatility, heavier put buying and a more bearish survey balance raise the MDI; a calm, bullish tape lowers it. The Sentiment WoE combines five measures of fear and positioning, each scored against its own trailing year:

• Bull/Bear Spread: Investor sentiment surveys.
• VIX (Level): The absolute price level of equity volatility.
• MOVE (Level): The absolute level of bond volatility.
• CBOE Index Put/Call Ratio: Hedging activity for broad indices.
• CBOE SPX Put/Call Ratio: Specific hedging activity for the S&P 500.

The Utility: This pillar registers fear as it builds. Rising hedging demand and bearish positioning are read as danger, not as a buying opportunity: the pillar is not contrarian: its task is to register danger while stress is rising, and it leaves bottom-picking to the allocator.

[MACRO HEALTH] (Secondary Weighting)

The Concept: Measures the absolute "oxygen" in the financial system. It tracks Net Systemic Liquidity (Central Bank balance sheets minus reverse repo and treasury general accounts), the Copper/Gold ratio (industrial demand vs safe haven), consumer discretionary against consumer staples shares, the oil price, the dollar index, and the 10-year/2-year Treasury spread measured against its own trailing year. Rising liquidity, a rising copper/gold ratio and discretionary shares outperforming staples lower the MDI; rising oil and a rising dollar raise it. A curve steepening sharply relative to its recent history raises the MDI: that is how an easing cycle into weakness shows up, and it accompanied the 2001 and 2007–08 bear markets. It is deliberately not an inversion signal — inversion has historically led recessions by a year or more, far beyond this pillar's horizon.

The Utility: It is the slow structural check. Even with sentiment complacent and volatility suppressed, a deteriorating Macro Health reading registers that the funding and growth conditions underneath are weakening.

6. Stability Mechanisms: Keeping the Reading Steady

A regime reading is only useful if it does not flicker. Without stability mechanisms, a strategy following the score would trade nearly every day and lose much of its value to commissions, slippage, and taxes.

1. The Vol-Cap Circuit Breaker (The 2022 Fix): In the reference strategy, if SPX realized volatility exceeds a fixed, extreme annualized threshold, exposure is cut whatever the regime: when the market moves too violently, the reliability of every indicator drops sharply. The terminal shows the cap's state as a reading on its first screen.
2. The 5-Day Temporal Consensus Filter: Instead of acting on a single daily data point, the model selects the Mode (most frequent regime) of a rolling 5-day window. A regime change is only confirmed if it is the "majority view" of the trading week; a week with two regimes tied at two sessions each resolves to the calmer of the two.
3. Threshold Decoupling (Hysteresis Buffer): To enter WARNING, the MDI must cross above a threshold (e.g., 55). To exit, it must drop significantly below a lower threshold (e.g., 45). This buffer zone filters out "Dead Cat Bounces."
4. Regime Conviction Score: An equal-weighted 4-factor composite of how settled the regime is (Zonal Penetration, Velocity Trajectory, Temporal Persistence, Markov Historical Retention: the regime's five-session retention in the published record, measured on the 2026-09-30 edition as CALM 79.5%, CHOP 81.3%, WARNING 82.3%, CRITICAL 63.0%), shown beside the regime with turnover sensitivity as a reading: HIGH below a score of 70. It is a reading, not an instruction or a setting; what a firm does with it is the firm's decision.
5. Institutional Analytics: HRP, Gerber Statistic, and Marchenko-Pastur Denoising: On the Risk Desk, a client’s book is measured against a Hierarchical Risk Parity (HRP) reference allocation rather than a Markowitz optimum. The HRP reference is fitted on a covariance estimated with the Gerber Statistic and, on books of ten or more holdings, cleansed with Marchenko-Pastur Denoising; the correlation matrix and the risk contributions are measured separately, on the sample covariance shrunk toward constant correlation (Ledoit and Wolf). It is a yardstick, not a target: Oneiros Vision supplies no target allocation, and the only target the terminal trades toward is the one the portfolio manager sets.
6. Implementation Research: How Fast the Illustrative Model Must Trade: The Implementation Sensitivity study capped the largest single-position move per session when the illustrative model changes regime. Across seven allocation ladders, a 10% cap keeps at least 98% of the drawdown benefit of trading the full rotation at the next close; at 7.5% the weakest ladder keeps 80%, and slower caps lose the benefit as the signal decays. The finding describes the illustrative model, not a client’s book: the terminal moves no book toward the model, and on the Execution screen turns the portfolio manager’s own target into orders at a pace the desk decides.

7. Explainability and Measured Uncertainty

A risk model that cannot say why it reads what it reads is a black box, and a number published without its uncertainty invites more confidence than its evidence supports. Two properties of the engine address this directly, and both are arithmetic rather than approximation.

7.1 The Index Decomposes Exactly

The Master Danger Index is affine in its three pillars, and the Speedometer is a mean of normal CDFs. Substituting the pillar definitions gives fixed neutral points of 20, 20 and 10, so the index is 50 plus the sum of signed input contributions.

This is an identity, not an estimate. The contributions sum to the score to the ninth decimal place on every session of the frozen historical record — 14,971 rows — and that equality is enforced as a hard test rather than a tolerance. “Why is the index at 74” therefore has an exact answer: each of the seventeen inputs (sixteen series: VIX and MOVE enter two pillars, consumer discretionary and staples as one ratio) carries a signed contribution in index points, rankable, with no sampling and no surrogate model. For explainability obligations this is a stronger position than a sampled attribution method, because it is the arithmetic of the score itself rather than an approximation of it.

Two properties are reported that a naive decomposition would omit. Volatility enters twice — once as vol-of-vol through the Speedometer and once as a level z-score through Sentiment — so both legs and their total are shown. And saturation is disclosed: the Sentiment and Macro pillars are bounded, and once a pillar reaches its bound it stops responding to its inputs entirely. Macro is bounded on roughly 41% of CRITICAL sessions, precisely when an explanation is most wanted, so contributions are reported before the bound is applied alongside the points the bound absorbed.

7.2 Uncertainty Is Published, Not Withheld

Regime transition probabilities are empirical base rates counted from the realised record, and their daily observation counts overstate their evidence substantially. A rate resting on roughly twelve hundred sessions may stand on only a few dozen independent regime episodes, because a regime holding for six weeks contributes thirty observations and one episode.

Every such rate is published with a 95% interval from a stationary block bootstrap — blocked at the mean regime episode length, so the resampling reflects how few independent episodes there are — and with the episode count beside the observation count. The same treatment is applied to the backtest's headline performance figures. This widens the stated uncertainty; it moves no published estimate.

8. Structural Dependency: Testing the Correlation 1.0 Thesis

Section 1 rests on the claim that conventional diversification failed in 2022. That claim is now measured rather than asserted, and measuring it produced a result worth stating precisely, because the obvious statistic is the wrong one.

Lower tail dependence does not evidence this thesis. It measures the probability two assets are in their worst days together. Measured on the S&P 500 against intermediate Treasuries through 2022 it reads 0.004 — very nearly zero. On that number the hedge appears intact, and it was not: bonds declined steadily on rate repricing rather than crashing alongside equities on the same sessions.

What changed was the sign of the relationship. The two correlated between −0.35 and −0.68 in every calendar year from 2010 through 2020. In 2022 the correlation was +0.19, both assets lost money over the year, and the share of sessions on which both declined rose from a range of 8–17% to 33.5%. That is the measurable content of a Correlation 1.0 event, and the platform reports it on a client's own holdings rather than as a general assertion.

The dependence structure is estimated with a t-copula rather than a Gaussian one. A Gaussian copula has zero tail dependence for any correlation below 1: it cannot express the property being tested, whatever the data shows. That limitation is what made it notorious in pre-2008 credit modelling, and using it here would repeat the error in a different asset class.

9. Where This Applies: Currency, Jurisdiction and What Transfers

Every preceding section measures an American signal on an American index in dollars. That is a real limit on what may be claimed elsewhere, and it is stated here rather than left for a prospect to discover.

9.1 The signal is American, and nothing here pretends otherwise

Every input to the Master Danger Index is a United States series: Federal Reserve net liquidity, the US high-yield spread, the US yield curve, VIX, MOVE, the dollar index, US sector ratios and US sentiment surveys. Two of them are the dollar’s crosses against the euro and the franc, which are dollar series read from the other side. There is no European, British or Asian input, the strategy holds US-listed instruments, and the chassis is frozen — none of that changes because a client is in Zurich, Frankfurt or London.

So the honest question from a non-US desk is not whether the signal works. It is whether an American signal says anything about their market, and whether the dollar figures are the ones they would actually have earned. Both have been measured.

9.2 What transfers, against criteria fixed in advance

The regime transfer study applies the same measurement behind the tear sheet’s Layer 1 figures to indices the engine has never seen. Four criteria were fixed in writing before the first run — a monotonic forward drawdown across regimes, at least 60% of the index’s worst 1% of sessions held by the defensive regimes on at most 40% of sessions, no reversal when restated in another currency, and at least 20 independent stress episodes — so that no result could be obtained by moving a threshold afterwards.

MSCI World did not meet the threshold and is reported as a failure. Its table is as monotonic as the others, but the instrument used to measure it lists from 2012 and yields 17 independent stress episodes where the rule requires 20. Not demonstrated is a different statement from does not work; the remedy is a longer series, not a lower threshold and not the same test re-run later hoping the count has grown.

The FTSE result has two halves and a UK reader is entitled to both. Its tail capture is the highest in the study — higher than the S&P 500 control — so on this evidence the regime identifies British stress at least as well as American stress. Its gradient, the CRITICAL forward drawdown over the CALM one, is the shallowest in the study: 1.42x against the control’s 1.99x. The FTSE is commodity- and staples-heavy and earns internationally, so it does not sell off as violently. Read together, the overlay’s value to a British book is concentrated in avoiding the worst sessions rather than in a large reduction of typical drawdown.

9.3 Currency is the larger number, and it is not skill

A client reporting in another currency did not earn the dollar figures in the tear sheet. They earned them compounded with what the dollar did against their own currency, and over this window that difference is larger than most of what separates the strategy from its benchmarks. The franc drawdown is the number that matters most here, and it is not a strategy drawdown — unhedged in francs the same track drew down 30.11% in 2011, of which 7.60 points were the strategy and the rest was the exchange rate. A Swiss client is entitled to see that figure before they are shown -13.61%, not after. Sterling ran the other way and flattered a pound-based return by 1.38 pp a year, which is the same effect with the sign reversed and no more repeatable. The full four-currency table is in the performance tear sheet.

9.4 What the regulation means, and whose obligation it is

Oneiros Vision measures. It places no orders, holds no client money or assets, exercises no discretion and makes no recommendation. That is a deliberate constraint on the product, not a gap in it, and it decides most of what follows.

The obligations below bind the client firm, not Oneiros Vision. What the platform does is produce the evidence a firm needs to discharge them: the 10% depreciation notice under MiFID II Article 62 of Commission Delegated Regulation (EU) 2017/565, onshored in the UK as COBS 16A.4.3R; liquidity stress testing under ESMA34-39-897; UK distribution status, which since Brexit is a different question from domicile; UK stamp duty at 0.5% on purchases of shares in UK-incorporated companies, stated as a transaction cost; and third-party due diligence under SYSC 8 in the UK and DORA in the EU.

9.5 What this section does not claim

It does not claim that Oneiros Vision is a European or British model. No European or British series is an input, the strategy holds US-listed instruments, and the chassis is frozen. The evidence above is that an American signal also ordered risk abroad — a statement about transfer, and nothing wider.

It does not claim regulatory compliance on a client’s behalf. Every obligation named above remains the firm’s. The platform supplies measurements a firm may cite; it does not certify anything and cannot be cited as a control. And it does not claim the currency effects are repeatable: they are what the dollar did over one window against three currencies, reported so that a client sees their own number rather than somebody else’s.

10. The API: the Terminal’s Measurements, Read by Your Own Systems

Some firms already run their own portfolio and reporting systems. The API plan gives them programmatic access to what the terminal measures, under the same rules the terminal follows.

• The regime reading: the Master Danger Index, the regime and the pillars behind it, each session’s reading and its history, for a firm’s own risk reports and monitoring.
• Portfolio measurements: a platform passes a portfolio and receives how it has behaved in each regime, its risk decomposition and its stress figures, measured on its own holdings and in its own currency.
• Access control: keys are issued per account, can be restricted to the caller’s own IP addresses, and every call is logged.

The API reports measurements. It returns no target allocation and no trade instruction: what a firm’s own systems do with a reading is the firm’s decision, and the firm’s obligation (§9).

11. Conclusion: A Disciplined Framework for Dynamic Risk Governance

Oneiros Vision replaces discretionary regime calls with a repeatable, rules-based reading that can be checked: the same inputs, the same arithmetic and the same thresholds every session, each call dated and kept. It gives a portfolio manager a measured account of the market's regime and of a book's exposure to it, with the uncertainty stated and the limits named.

The architecture documented in this whitepaper — from the pillar telemetry to the stability mechanisms — exists to make that reading trustworthy: stable enough not to flicker, transparent enough to be checked line by line, and plain about what it does not know. What a firm does with it remains the firm's decision.