THE capstone · all validated sleeves at frozen 439 weights · MC base ~19.5%/yr · excluded from aggregate

everything RH agentic can execute · 439c max-return mix · MC base ~13.1%/yr · excluded from aggregate

ML K5/H60 with a standing 0.30-unit SPY short. Halves beta 0.61→0.31 and cuts worst-day losses ~60%. Inherits the parent's label-leakage correction: purged Sharpe 0.76, not 1.99.

ML Liquidity with a standing 0.30-unit SPY short. Same caveats as the parent — passed its control and cost gate, failed transfer and DSR.

Long announced cash-merger targets · deal-break-insurance spread · beta ~0.2 · 20 slots · paper from 2026-07-27

XGB ensemble · 78 features · 5d hold

SEC Form-4 insider buying · top-3 · 60d hold w/ 4-stage scale-out

3-factor L/S · K=10 · 60d hold · automated

5-factor L/S · K=3 · 60d hold · tradable (6/day)

5-sleeve Ultimate: GLD/EFA + sector-rot + LEAPS L/S + Low-Vol + hedge · BT Sh +3.50

Momentum + quality + short-term-reversal · market-neutral L/S · shadow

Cross-asset trend (bonds+commodities, equity-stripped) · 3/6/12-mo ensemble · monthly · uncorrelated diversifier

Delta-neutral BTC+ETH long-spot/short-perp · collect funding · SPY-uncorrelated carry

90-DTE ATM calls mirroring live insider entries · v2 rule 07-21 after the 45-DTE expiry-accounting artifact was caught

ML ticket-outcome model · top-10 condors/mo · abstains when no edge · frozen v1

Systematic short SPX variance · defined-risk iron condors · regime-gated · return sleeve

Long basket of 8-K share-repurchase announcers · tracking-error alpha vs SPY

Market-neutral RSI + momentum-acceleration long/short. Beta -0.04. The 1.50 hedged Sharpe it was promoted on was a window artifact — honest full-panel figure is +0.69, and +0.05 across 2022-23.

Same 78 ml features, K=5 and a 60-day hold. The 1.99 Sharpe it was built on was LABEL LEAKAGE — purged, it is 0.76, and the edge over the deployed cell is smaller than seed noise. Forward record is valid; the advertised number never was.

Base 78 features plus the 12 liquidity columns — the only family of ~380 never-merged reference columns to pass a random-block control (z=3.72) and survive costs flat to 100bp. It FAILED cross-model transfer and DSR, so this is a forward test, not a promotion.