BriefRooms Research

BRACE-SPX Lab

A laboratory for robust S&P 500 exposure management. We publish research results, governance and shadow status. Code, parameters, raw forecasts and the full ledger remain private.

Loading the sanitized research report…

Laboratory status

—

— — —

Current architecture

—

Candidate signature—immutable hash of ten constructions

FrequencyDaily / Weekly—

Development—holdout excluded from downloads

Five distinct sourcesTrend, breadth, liquidity, options/VIX and market-implied rates.
Higher frequencyDaily signals, scheduled weekly rebalance and a daily shock gate.
Deterministic regimesLogic changes across calm, high-volatility and panic conditions.
Global multiple testingThe assessment includes all prior and current trials.

Architecture 2 development gate

—

PBO—threshold ≤20%; lower is better

Global DSR—after correcting the complete trial history

Total trials—Generations 1–5 plus Architecture 2

Rank stability—median fold rank correlation

Fold winners—distinct leaders across six folds

External validation—independent Backtrader implementation

Selection—a single leader is prohibited

Holdout—not downloaded or opened

Important: PBO fell below 20%, but that was not enough. The cumulative 644-trial correction and unstable ranking block promotion of a single model.

Best diagnostic variant — not a champion

These figures describe the strongest development result. It is not authorized for the holdout or live use.

CAGR—annual growth rate

Volatility—annualized

Excess Sharpe—above the risk-free rate

Max drawdown—deepest equity decline

Calmar—CAGR / drawdown

Annual turnover—after exposure changes

Positive folds—chronological validation

P(CAGR advantage >0)—bootstrap versus Buy & Hold

P(Sharpe advantage >0)—bootstrap versus Buy & Hold

Benchmark comparison

Method	CAGR	Volatility	Excess Sharpe	Max drawdown	Calmar	Turnover

Are the constructions genuinely different?

Median signal correlation—before the orthogonality audit

Effective signal count—from five named sources

Effective exposure count—diversity of position paths

Median return correlation—one instrument still aligns results

Signal orthogonality audit

Sources were selected without SPY returns, Sharpe, CAGR or the holdout. The criterion was informational independence and a minimum unique-variance contribution.

Effective rank—independence of five raw sources

Median correlation—among raw sub-signals

Components for 85%—main information axes

Trend and breadth remain useful context filters, but neither cleared the 15% unique-variance threshold as an independent source for the next experiment.

Prospective shadow tracking

—

All constructions are tracked in parallel. Shadow mode places no orders, selects no champion and does not use the historical holdout to warm the signals.

— sessions — — remaining

Start—first session after the holdout ends

Orders—hypothetical records only

Champion—ten constructions tracked in parallel

Cold start252 sessionsbefore full signal-quality assessment

Sealed holdout

—

Architecture 2 downloaded only data ending before the holdout. Shadow tracking starts after it ends and creates a new prospective history.

Publication boundary

The browser receives only a sanitized JSON report. The private research engine is not part of the website code.

Research conclusion

—